]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 | 2 | /* |
d352ac68 | 3 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
4 | */ |
5 | ||
a6b6e75e | 6 | #include <linux/sched.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
bd989ba3 | 8 | #include <linux/rbtree.h> |
adf02123 | 9 | #include <linux/mm.h> |
eb60ceac CM |
10 | #include "ctree.h" |
11 | #include "disk-io.h" | |
7f5c1516 | 12 | #include "transaction.h" |
5f39d397 | 13 | #include "print-tree.h" |
925baedd | 14 | #include "locking.h" |
de37aa51 | 15 | #include "volumes.h" |
f616f5cd | 16 | #include "qgroup.h" |
9a8dd150 | 17 | |
e089f05c CM |
18 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
19 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
20 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
21 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
22 | int data_size, int extend); | |
5f39d397 | 23 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 24 | struct extent_buffer *dst, |
971a1f66 | 25 | struct extent_buffer *src, int empty); |
5f39d397 | 26 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
27 | struct extent_buffer *dst_buf, |
28 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
29 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
30 | int level, int slot); | |
d97e63b6 | 31 | |
df24a2b9 | 32 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 33 | { |
e2c89907 | 34 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
35 | } |
36 | ||
b4ce94de CM |
37 | /* |
38 | * set all locked nodes in the path to blocking locks. This should | |
39 | * be done before scheduling | |
40 | */ | |
41 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
42 | { | |
43 | int i; | |
44 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
45 | if (!p->nodes[i] || !p->locks[i]) |
46 | continue; | |
766ece54 DS |
47 | /* |
48 | * If we currently have a spinning reader or writer lock this | |
49 | * will bump the count of blocking holders and drop the | |
50 | * spinlock. | |
51 | */ | |
52 | if (p->locks[i] == BTRFS_READ_LOCK) { | |
53 | btrfs_set_lock_blocking_read(p->nodes[i]); | |
bd681513 | 54 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; |
766ece54 DS |
55 | } else if (p->locks[i] == BTRFS_WRITE_LOCK) { |
56 | btrfs_set_lock_blocking_write(p->nodes[i]); | |
bd681513 | 57 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; |
766ece54 | 58 | } |
b4ce94de CM |
59 | } |
60 | } | |
61 | ||
d352ac68 | 62 | /* this also releases the path */ |
df24a2b9 | 63 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 64 | { |
ff175d57 JJ |
65 | if (!p) |
66 | return; | |
b3b4aa74 | 67 | btrfs_release_path(p); |
df24a2b9 | 68 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
69 | } |
70 | ||
d352ac68 CM |
71 | /* |
72 | * path release drops references on the extent buffers in the path | |
73 | * and it drops any locks held by this path | |
74 | * | |
75 | * It is safe to call this on paths that no locks or extent buffers held. | |
76 | */ | |
b3b4aa74 | 77 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
78 | { |
79 | int i; | |
a2135011 | 80 | |
234b63a0 | 81 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 82 | p->slots[i] = 0; |
eb60ceac | 83 | if (!p->nodes[i]) |
925baedd CM |
84 | continue; |
85 | if (p->locks[i]) { | |
bd681513 | 86 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
87 | p->locks[i] = 0; |
88 | } | |
5f39d397 | 89 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 90 | p->nodes[i] = NULL; |
eb60ceac CM |
91 | } |
92 | } | |
93 | ||
d352ac68 CM |
94 | /* |
95 | * safely gets a reference on the root node of a tree. A lock | |
96 | * is not taken, so a concurrent writer may put a different node | |
97 | * at the root of the tree. See btrfs_lock_root_node for the | |
98 | * looping required. | |
99 | * | |
100 | * The extent buffer returned by this has a reference taken, so | |
101 | * it won't disappear. It may stop being the root of the tree | |
102 | * at any time because there are no locks held. | |
103 | */ | |
925baedd CM |
104 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
105 | { | |
106 | struct extent_buffer *eb; | |
240f62c8 | 107 | |
3083ee2e JB |
108 | while (1) { |
109 | rcu_read_lock(); | |
110 | eb = rcu_dereference(root->node); | |
111 | ||
112 | /* | |
113 | * RCU really hurts here, we could free up the root node because | |
01327610 | 114 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
115 | * the inc_not_zero dance and if it doesn't work then |
116 | * synchronize_rcu and try again. | |
117 | */ | |
118 | if (atomic_inc_not_zero(&eb->refs)) { | |
119 | rcu_read_unlock(); | |
120 | break; | |
121 | } | |
122 | rcu_read_unlock(); | |
123 | synchronize_rcu(); | |
124 | } | |
925baedd CM |
125 | return eb; |
126 | } | |
127 | ||
d352ac68 CM |
128 | /* loop around taking references on and locking the root node of the |
129 | * tree until you end up with a lock on the root. A locked buffer | |
130 | * is returned, with a reference held. | |
131 | */ | |
925baedd CM |
132 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
133 | { | |
134 | struct extent_buffer *eb; | |
135 | ||
d397712b | 136 | while (1) { |
925baedd CM |
137 | eb = btrfs_root_node(root); |
138 | btrfs_tree_lock(eb); | |
240f62c8 | 139 | if (eb == root->node) |
925baedd | 140 | break; |
925baedd CM |
141 | btrfs_tree_unlock(eb); |
142 | free_extent_buffer(eb); | |
143 | } | |
144 | return eb; | |
145 | } | |
146 | ||
bd681513 CM |
147 | /* loop around taking references on and locking the root node of the |
148 | * tree until you end up with a lock on the root. A locked buffer | |
149 | * is returned, with a reference held. | |
150 | */ | |
84f7d8e6 | 151 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
152 | { |
153 | struct extent_buffer *eb; | |
154 | ||
155 | while (1) { | |
156 | eb = btrfs_root_node(root); | |
157 | btrfs_tree_read_lock(eb); | |
158 | if (eb == root->node) | |
159 | break; | |
160 | btrfs_tree_read_unlock(eb); | |
161 | free_extent_buffer(eb); | |
162 | } | |
163 | return eb; | |
164 | } | |
165 | ||
d352ac68 CM |
166 | /* cowonly root (everything not a reference counted cow subvolume), just get |
167 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
168 | * get properly updated on disk. | |
169 | */ | |
0b86a832 CM |
170 | static void add_root_to_dirty_list(struct btrfs_root *root) |
171 | { | |
0b246afa JM |
172 | struct btrfs_fs_info *fs_info = root->fs_info; |
173 | ||
e7070be1 JB |
174 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
175 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
176 | return; | |
177 | ||
0b246afa | 178 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
179 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
180 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 181 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 182 | list_move_tail(&root->dirty_list, |
0b246afa | 183 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
184 | else |
185 | list_move(&root->dirty_list, | |
0b246afa | 186 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 187 | } |
0b246afa | 188 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
189 | } |
190 | ||
d352ac68 CM |
191 | /* |
192 | * used by snapshot creation to make a copy of a root for a tree with | |
193 | * a given objectid. The buffer with the new root node is returned in | |
194 | * cow_ret, and this func returns zero on success or a negative error code. | |
195 | */ | |
be20aa9d CM |
196 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
197 | struct btrfs_root *root, | |
198 | struct extent_buffer *buf, | |
199 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
200 | { | |
0b246afa | 201 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 202 | struct extent_buffer *cow; |
be20aa9d CM |
203 | int ret = 0; |
204 | int level; | |
5d4f98a2 | 205 | struct btrfs_disk_key disk_key; |
be20aa9d | 206 | |
27cdeb70 | 207 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 208 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
209 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
210 | trans->transid != root->last_trans); | |
be20aa9d CM |
211 | |
212 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
213 | if (level == 0) |
214 | btrfs_item_key(buf, &disk_key, 0); | |
215 | else | |
216 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 217 | |
4d75f8a9 DS |
218 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
219 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 220 | if (IS_ERR(cow)) |
be20aa9d CM |
221 | return PTR_ERR(cow); |
222 | ||
58e8012c | 223 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
224 | btrfs_set_header_bytenr(cow, cow->start); |
225 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
226 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
227 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
228 | BTRFS_HEADER_FLAG_RELOC); | |
229 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
230 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
231 | else | |
232 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 233 | |
de37aa51 | 234 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 235 | |
be20aa9d | 236 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 237 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 238 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 239 | else |
e339a6b0 | 240 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 241 | |
be20aa9d CM |
242 | if (ret) |
243 | return ret; | |
244 | ||
245 | btrfs_mark_buffer_dirty(cow); | |
246 | *cow_ret = cow; | |
247 | return 0; | |
248 | } | |
249 | ||
bd989ba3 JS |
250 | enum mod_log_op { |
251 | MOD_LOG_KEY_REPLACE, | |
252 | MOD_LOG_KEY_ADD, | |
253 | MOD_LOG_KEY_REMOVE, | |
254 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
255 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
256 | MOD_LOG_MOVE_KEYS, | |
257 | MOD_LOG_ROOT_REPLACE, | |
258 | }; | |
259 | ||
bd989ba3 JS |
260 | struct tree_mod_root { |
261 | u64 logical; | |
262 | u8 level; | |
263 | }; | |
264 | ||
265 | struct tree_mod_elem { | |
266 | struct rb_node node; | |
298cfd36 | 267 | u64 logical; |
097b8a7c | 268 | u64 seq; |
bd989ba3 JS |
269 | enum mod_log_op op; |
270 | ||
271 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
272 | int slot; | |
273 | ||
274 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
275 | u64 generation; | |
276 | ||
277 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
278 | struct btrfs_disk_key key; | |
279 | u64 blockptr; | |
280 | ||
281 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
b6dfa35b DS |
282 | struct { |
283 | int dst_slot; | |
284 | int nr_items; | |
285 | } move; | |
bd989ba3 JS |
286 | |
287 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
288 | struct tree_mod_root old_root; | |
289 | }; | |
290 | ||
fc36ed7e | 291 | /* |
fcebe456 | 292 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 293 | */ |
fcebe456 | 294 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
295 | { |
296 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
297 | } | |
298 | ||
097b8a7c JS |
299 | /* |
300 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
301 | * passed does not already have a sequence number set. So when a caller expects | |
302 | * to record tree modifications, it should ensure to set elem->seq to zero | |
303 | * before calling btrfs_get_tree_mod_seq. | |
304 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
305 | * blocker was added. | |
306 | */ | |
307 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
308 | struct seq_list *elem) | |
bd989ba3 | 309 | { |
b1a09f1e | 310 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 | 311 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 312 | if (!elem->seq) { |
fcebe456 | 313 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
314 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
315 | } | |
bd989ba3 | 316 | spin_unlock(&fs_info->tree_mod_seq_lock); |
b1a09f1e | 317 | write_unlock(&fs_info->tree_mod_log_lock); |
097b8a7c | 318 | |
fcebe456 | 319 | return elem->seq; |
bd989ba3 JS |
320 | } |
321 | ||
322 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
323 | struct seq_list *elem) | |
324 | { | |
325 | struct rb_root *tm_root; | |
326 | struct rb_node *node; | |
327 | struct rb_node *next; | |
328 | struct seq_list *cur_elem; | |
329 | struct tree_mod_elem *tm; | |
330 | u64 min_seq = (u64)-1; | |
331 | u64 seq_putting = elem->seq; | |
332 | ||
333 | if (!seq_putting) | |
334 | return; | |
335 | ||
bd989ba3 JS |
336 | spin_lock(&fs_info->tree_mod_seq_lock); |
337 | list_del(&elem->list); | |
097b8a7c | 338 | elem->seq = 0; |
bd989ba3 JS |
339 | |
340 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 341 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
342 | if (seq_putting > cur_elem->seq) { |
343 | /* | |
344 | * blocker with lower sequence number exists, we | |
345 | * cannot remove anything from the log | |
346 | */ | |
097b8a7c JS |
347 | spin_unlock(&fs_info->tree_mod_seq_lock); |
348 | return; | |
bd989ba3 JS |
349 | } |
350 | min_seq = cur_elem->seq; | |
351 | } | |
352 | } | |
097b8a7c JS |
353 | spin_unlock(&fs_info->tree_mod_seq_lock); |
354 | ||
bd989ba3 JS |
355 | /* |
356 | * anything that's lower than the lowest existing (read: blocked) | |
357 | * sequence number can be removed from the tree. | |
358 | */ | |
b1a09f1e | 359 | write_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
360 | tm_root = &fs_info->tree_mod_log; |
361 | for (node = rb_first(tm_root); node; node = next) { | |
362 | next = rb_next(node); | |
6b4df8b6 | 363 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 364 | if (tm->seq > min_seq) |
bd989ba3 JS |
365 | continue; |
366 | rb_erase(node, tm_root); | |
bd989ba3 JS |
367 | kfree(tm); |
368 | } | |
b1a09f1e | 369 | write_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
370 | } |
371 | ||
372 | /* | |
373 | * key order of the log: | |
298cfd36 | 374 | * node/leaf start address -> sequence |
bd989ba3 | 375 | * |
298cfd36 CR |
376 | * The 'start address' is the logical address of the *new* root node |
377 | * for root replace operations, or the logical address of the affected | |
378 | * block for all other operations. | |
bd989ba3 JS |
379 | */ |
380 | static noinline int | |
381 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
382 | { | |
383 | struct rb_root *tm_root; | |
384 | struct rb_node **new; | |
385 | struct rb_node *parent = NULL; | |
386 | struct tree_mod_elem *cur; | |
c8cc6341 | 387 | |
73e82fe4 DS |
388 | lockdep_assert_held_write(&fs_info->tree_mod_log_lock); |
389 | ||
fcebe456 | 390 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 391 | |
bd989ba3 JS |
392 | tm_root = &fs_info->tree_mod_log; |
393 | new = &tm_root->rb_node; | |
394 | while (*new) { | |
6b4df8b6 | 395 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 396 | parent = *new; |
298cfd36 | 397 | if (cur->logical < tm->logical) |
bd989ba3 | 398 | new = &((*new)->rb_left); |
298cfd36 | 399 | else if (cur->logical > tm->logical) |
bd989ba3 | 400 | new = &((*new)->rb_right); |
097b8a7c | 401 | else if (cur->seq < tm->seq) |
bd989ba3 | 402 | new = &((*new)->rb_left); |
097b8a7c | 403 | else if (cur->seq > tm->seq) |
bd989ba3 | 404 | new = &((*new)->rb_right); |
5de865ee FDBM |
405 | else |
406 | return -EEXIST; | |
bd989ba3 JS |
407 | } |
408 | ||
409 | rb_link_node(&tm->node, parent, new); | |
410 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 411 | return 0; |
bd989ba3 JS |
412 | } |
413 | ||
097b8a7c JS |
414 | /* |
415 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
416 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
417 | * this until all tree mod log insertions are recorded in the rb tree and then | |
b1a09f1e | 418 | * write unlock fs_info::tree_mod_log_lock. |
097b8a7c | 419 | */ |
e9b7fd4d JS |
420 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
421 | struct extent_buffer *eb) { | |
422 | smp_mb(); | |
423 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
424 | return 1; | |
097b8a7c JS |
425 | if (eb && btrfs_header_level(eb) == 0) |
426 | return 1; | |
5de865ee | 427 | |
b1a09f1e | 428 | write_lock(&fs_info->tree_mod_log_lock); |
5de865ee | 429 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { |
b1a09f1e | 430 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
431 | return 1; |
432 | } | |
433 | ||
e9b7fd4d JS |
434 | return 0; |
435 | } | |
436 | ||
5de865ee FDBM |
437 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
438 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
439 | struct extent_buffer *eb) | |
440 | { | |
441 | smp_mb(); | |
442 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
443 | return 0; | |
444 | if (eb && btrfs_header_level(eb) == 0) | |
445 | return 0; | |
446 | ||
447 | return 1; | |
448 | } | |
449 | ||
450 | static struct tree_mod_elem * | |
451 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
452 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 453 | { |
097b8a7c | 454 | struct tree_mod_elem *tm; |
bd989ba3 | 455 | |
c8cc6341 JB |
456 | tm = kzalloc(sizeof(*tm), flags); |
457 | if (!tm) | |
5de865ee | 458 | return NULL; |
bd989ba3 | 459 | |
298cfd36 | 460 | tm->logical = eb->start; |
bd989ba3 JS |
461 | if (op != MOD_LOG_KEY_ADD) { |
462 | btrfs_node_key(eb, &tm->key, slot); | |
463 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
464 | } | |
465 | tm->op = op; | |
466 | tm->slot = slot; | |
467 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 468 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 469 | |
5de865ee | 470 | return tm; |
097b8a7c JS |
471 | } |
472 | ||
e09c2efe DS |
473 | static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot, |
474 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 475 | { |
5de865ee FDBM |
476 | struct tree_mod_elem *tm; |
477 | int ret; | |
478 | ||
e09c2efe | 479 | if (!tree_mod_need_log(eb->fs_info, eb)) |
5de865ee FDBM |
480 | return 0; |
481 | ||
482 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
483 | if (!tm) | |
484 | return -ENOMEM; | |
485 | ||
e09c2efe | 486 | if (tree_mod_dont_log(eb->fs_info, eb)) { |
5de865ee | 487 | kfree(tm); |
097b8a7c | 488 | return 0; |
5de865ee FDBM |
489 | } |
490 | ||
e09c2efe | 491 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
b1a09f1e | 492 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
493 | if (ret) |
494 | kfree(tm); | |
097b8a7c | 495 | |
5de865ee | 496 | return ret; |
097b8a7c JS |
497 | } |
498 | ||
6074d45f DS |
499 | static noinline int tree_mod_log_insert_move(struct extent_buffer *eb, |
500 | int dst_slot, int src_slot, int nr_items) | |
bd989ba3 | 501 | { |
5de865ee FDBM |
502 | struct tree_mod_elem *tm = NULL; |
503 | struct tree_mod_elem **tm_list = NULL; | |
504 | int ret = 0; | |
bd989ba3 | 505 | int i; |
5de865ee | 506 | int locked = 0; |
bd989ba3 | 507 | |
6074d45f | 508 | if (!tree_mod_need_log(eb->fs_info, eb)) |
f395694c | 509 | return 0; |
bd989ba3 | 510 | |
176ef8f5 | 511 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
512 | if (!tm_list) |
513 | return -ENOMEM; | |
514 | ||
176ef8f5 | 515 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
516 | if (!tm) { |
517 | ret = -ENOMEM; | |
518 | goto free_tms; | |
519 | } | |
520 | ||
298cfd36 | 521 | tm->logical = eb->start; |
5de865ee FDBM |
522 | tm->slot = src_slot; |
523 | tm->move.dst_slot = dst_slot; | |
524 | tm->move.nr_items = nr_items; | |
525 | tm->op = MOD_LOG_MOVE_KEYS; | |
526 | ||
527 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
528 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 529 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
530 | if (!tm_list[i]) { |
531 | ret = -ENOMEM; | |
532 | goto free_tms; | |
533 | } | |
534 | } | |
535 | ||
6074d45f | 536 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
537 | goto free_tms; |
538 | locked = 1; | |
539 | ||
01763a2e JS |
540 | /* |
541 | * When we override something during the move, we log these removals. | |
542 | * This can only happen when we move towards the beginning of the | |
543 | * buffer, i.e. dst_slot < src_slot. | |
544 | */ | |
bd989ba3 | 545 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
6074d45f | 546 | ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]); |
5de865ee FDBM |
547 | if (ret) |
548 | goto free_tms; | |
bd989ba3 JS |
549 | } |
550 | ||
6074d45f | 551 | ret = __tree_mod_log_insert(eb->fs_info, tm); |
5de865ee FDBM |
552 | if (ret) |
553 | goto free_tms; | |
b1a09f1e | 554 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee | 555 | kfree(tm_list); |
f395694c | 556 | |
5de865ee FDBM |
557 | return 0; |
558 | free_tms: | |
559 | for (i = 0; i < nr_items; i++) { | |
560 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
6074d45f | 561 | rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log); |
5de865ee FDBM |
562 | kfree(tm_list[i]); |
563 | } | |
564 | if (locked) | |
b1a09f1e | 565 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
566 | kfree(tm_list); |
567 | kfree(tm); | |
bd989ba3 | 568 | |
5de865ee | 569 | return ret; |
bd989ba3 JS |
570 | } |
571 | ||
5de865ee FDBM |
572 | static inline int |
573 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
574 | struct tree_mod_elem **tm_list, | |
575 | int nritems) | |
097b8a7c | 576 | { |
5de865ee | 577 | int i, j; |
097b8a7c JS |
578 | int ret; |
579 | ||
097b8a7c | 580 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
581 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
582 | if (ret) { | |
583 | for (j = nritems - 1; j > i; j--) | |
584 | rb_erase(&tm_list[j]->node, | |
585 | &fs_info->tree_mod_log); | |
586 | return ret; | |
587 | } | |
097b8a7c | 588 | } |
5de865ee FDBM |
589 | |
590 | return 0; | |
097b8a7c JS |
591 | } |
592 | ||
95b757c1 DS |
593 | static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root, |
594 | struct extent_buffer *new_root, int log_removal) | |
bd989ba3 | 595 | { |
95b757c1 | 596 | struct btrfs_fs_info *fs_info = old_root->fs_info; |
5de865ee FDBM |
597 | struct tree_mod_elem *tm = NULL; |
598 | struct tree_mod_elem **tm_list = NULL; | |
599 | int nritems = 0; | |
600 | int ret = 0; | |
601 | int i; | |
bd989ba3 | 602 | |
5de865ee | 603 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
604 | return 0; |
605 | ||
5de865ee FDBM |
606 | if (log_removal && btrfs_header_level(old_root) > 0) { |
607 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 608 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 609 | GFP_NOFS); |
5de865ee FDBM |
610 | if (!tm_list) { |
611 | ret = -ENOMEM; | |
612 | goto free_tms; | |
613 | } | |
614 | for (i = 0; i < nritems; i++) { | |
615 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 616 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
617 | if (!tm_list[i]) { |
618 | ret = -ENOMEM; | |
619 | goto free_tms; | |
620 | } | |
621 | } | |
622 | } | |
d9abbf1c | 623 | |
bcc8e07f | 624 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
625 | if (!tm) { |
626 | ret = -ENOMEM; | |
627 | goto free_tms; | |
628 | } | |
bd989ba3 | 629 | |
298cfd36 | 630 | tm->logical = new_root->start; |
bd989ba3 JS |
631 | tm->old_root.logical = old_root->start; |
632 | tm->old_root.level = btrfs_header_level(old_root); | |
633 | tm->generation = btrfs_header_generation(old_root); | |
634 | tm->op = MOD_LOG_ROOT_REPLACE; | |
635 | ||
5de865ee FDBM |
636 | if (tree_mod_dont_log(fs_info, NULL)) |
637 | goto free_tms; | |
638 | ||
639 | if (tm_list) | |
640 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
641 | if (!ret) | |
642 | ret = __tree_mod_log_insert(fs_info, tm); | |
643 | ||
b1a09f1e | 644 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
645 | if (ret) |
646 | goto free_tms; | |
647 | kfree(tm_list); | |
648 | ||
649 | return ret; | |
650 | ||
651 | free_tms: | |
652 | if (tm_list) { | |
653 | for (i = 0; i < nritems; i++) | |
654 | kfree(tm_list[i]); | |
655 | kfree(tm_list); | |
656 | } | |
657 | kfree(tm); | |
658 | ||
659 | return ret; | |
bd989ba3 JS |
660 | } |
661 | ||
662 | static struct tree_mod_elem * | |
663 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
664 | int smallest) | |
665 | { | |
666 | struct rb_root *tm_root; | |
667 | struct rb_node *node; | |
668 | struct tree_mod_elem *cur = NULL; | |
669 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 670 | |
b1a09f1e | 671 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
672 | tm_root = &fs_info->tree_mod_log; |
673 | node = tm_root->rb_node; | |
674 | while (node) { | |
6b4df8b6 | 675 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 676 | if (cur->logical < start) { |
bd989ba3 | 677 | node = node->rb_left; |
298cfd36 | 678 | } else if (cur->logical > start) { |
bd989ba3 | 679 | node = node->rb_right; |
097b8a7c | 680 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
681 | node = node->rb_left; |
682 | } else if (!smallest) { | |
683 | /* we want the node with the highest seq */ | |
684 | if (found) | |
097b8a7c | 685 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
686 | found = cur; |
687 | node = node->rb_left; | |
097b8a7c | 688 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
689 | /* we want the node with the smallest seq */ |
690 | if (found) | |
097b8a7c | 691 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
692 | found = cur; |
693 | node = node->rb_right; | |
694 | } else { | |
695 | found = cur; | |
696 | break; | |
697 | } | |
698 | } | |
b1a09f1e | 699 | read_unlock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
700 | |
701 | return found; | |
702 | } | |
703 | ||
704 | /* | |
705 | * this returns the element from the log with the smallest time sequence | |
706 | * value that's in the log (the oldest log item). any element with a time | |
707 | * sequence lower than min_seq will be ignored. | |
708 | */ | |
709 | static struct tree_mod_elem * | |
710 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
711 | u64 min_seq) | |
712 | { | |
713 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
714 | } | |
715 | ||
716 | /* | |
717 | * this returns the element from the log with the largest time sequence | |
718 | * value that's in the log (the most recent log item). any element with | |
719 | * a time sequence lower than min_seq will be ignored. | |
720 | */ | |
721 | static struct tree_mod_elem * | |
722 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
723 | { | |
724 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
725 | } | |
726 | ||
ed874f0d | 727 | static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst, |
bd989ba3 | 728 | struct extent_buffer *src, unsigned long dst_offset, |
90f8d62e | 729 | unsigned long src_offset, int nr_items) |
bd989ba3 | 730 | { |
ed874f0d | 731 | struct btrfs_fs_info *fs_info = dst->fs_info; |
5de865ee FDBM |
732 | int ret = 0; |
733 | struct tree_mod_elem **tm_list = NULL; | |
734 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 735 | int i; |
5de865ee | 736 | int locked = 0; |
bd989ba3 | 737 | |
5de865ee FDBM |
738 | if (!tree_mod_need_log(fs_info, NULL)) |
739 | return 0; | |
bd989ba3 | 740 | |
c8cc6341 | 741 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
742 | return 0; |
743 | ||
31e818fe | 744 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
745 | GFP_NOFS); |
746 | if (!tm_list) | |
747 | return -ENOMEM; | |
bd989ba3 | 748 | |
5de865ee FDBM |
749 | tm_list_add = tm_list; |
750 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 751 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
752 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
753 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
754 | if (!tm_list_rem[i]) { | |
755 | ret = -ENOMEM; | |
756 | goto free_tms; | |
757 | } | |
758 | ||
759 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
760 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
761 | if (!tm_list_add[i]) { | |
762 | ret = -ENOMEM; | |
763 | goto free_tms; | |
764 | } | |
765 | } | |
766 | ||
767 | if (tree_mod_dont_log(fs_info, NULL)) | |
768 | goto free_tms; | |
769 | locked = 1; | |
770 | ||
771 | for (i = 0; i < nr_items; i++) { | |
772 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
773 | if (ret) | |
774 | goto free_tms; | |
775 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
776 | if (ret) | |
777 | goto free_tms; | |
bd989ba3 | 778 | } |
5de865ee | 779 | |
b1a09f1e | 780 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
781 | kfree(tm_list); |
782 | ||
783 | return 0; | |
784 | ||
785 | free_tms: | |
786 | for (i = 0; i < nr_items * 2; i++) { | |
787 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
788 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
789 | kfree(tm_list[i]); | |
790 | } | |
791 | if (locked) | |
b1a09f1e | 792 | write_unlock(&fs_info->tree_mod_log_lock); |
5de865ee FDBM |
793 | kfree(tm_list); |
794 | ||
795 | return ret; | |
bd989ba3 JS |
796 | } |
797 | ||
db7279a2 | 798 | static noinline int tree_mod_log_free_eb(struct extent_buffer *eb) |
bd989ba3 | 799 | { |
5de865ee FDBM |
800 | struct tree_mod_elem **tm_list = NULL; |
801 | int nritems = 0; | |
802 | int i; | |
803 | int ret = 0; | |
804 | ||
805 | if (btrfs_header_level(eb) == 0) | |
806 | return 0; | |
807 | ||
db7279a2 | 808 | if (!tree_mod_need_log(eb->fs_info, NULL)) |
5de865ee FDBM |
809 | return 0; |
810 | ||
811 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 812 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
813 | if (!tm_list) |
814 | return -ENOMEM; | |
815 | ||
816 | for (i = 0; i < nritems; i++) { | |
817 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
818 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
819 | if (!tm_list[i]) { | |
820 | ret = -ENOMEM; | |
821 | goto free_tms; | |
822 | } | |
823 | } | |
824 | ||
db7279a2 | 825 | if (tree_mod_dont_log(eb->fs_info, eb)) |
5de865ee FDBM |
826 | goto free_tms; |
827 | ||
db7279a2 | 828 | ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems); |
b1a09f1e | 829 | write_unlock(&eb->fs_info->tree_mod_log_lock); |
5de865ee FDBM |
830 | if (ret) |
831 | goto free_tms; | |
832 | kfree(tm_list); | |
833 | ||
834 | return 0; | |
835 | ||
836 | free_tms: | |
837 | for (i = 0; i < nritems; i++) | |
838 | kfree(tm_list[i]); | |
839 | kfree(tm_list); | |
840 | ||
841 | return ret; | |
bd989ba3 JS |
842 | } |
843 | ||
5d4f98a2 YZ |
844 | /* |
845 | * check if the tree block can be shared by multiple trees | |
846 | */ | |
847 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
848 | struct extent_buffer *buf) | |
849 | { | |
850 | /* | |
01327610 | 851 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
852 | * are never shared. If a block was allocated after the last |
853 | * snapshot and the block was not allocated by tree relocation, | |
854 | * we know the block is not shared. | |
855 | */ | |
27cdeb70 | 856 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
857 | buf != root->node && buf != root->commit_root && |
858 | (btrfs_header_generation(buf) <= | |
859 | btrfs_root_last_snapshot(&root->root_item) || | |
860 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
861 | return 1; | |
a79865c6 | 862 | |
5d4f98a2 YZ |
863 | return 0; |
864 | } | |
865 | ||
866 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
867 | struct btrfs_root *root, | |
868 | struct extent_buffer *buf, | |
f0486c68 YZ |
869 | struct extent_buffer *cow, |
870 | int *last_ref) | |
5d4f98a2 | 871 | { |
0b246afa | 872 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
873 | u64 refs; |
874 | u64 owner; | |
875 | u64 flags; | |
876 | u64 new_flags = 0; | |
877 | int ret; | |
878 | ||
879 | /* | |
880 | * Backrefs update rules: | |
881 | * | |
882 | * Always use full backrefs for extent pointers in tree block | |
883 | * allocated by tree relocation. | |
884 | * | |
885 | * If a shared tree block is no longer referenced by its owner | |
886 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
887 | * use full backrefs for extent pointers in tree block. | |
888 | * | |
889 | * If a tree block is been relocating | |
890 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
891 | * use full backrefs for extent pointers in tree block. | |
892 | * The reason for this is some operations (such as drop tree) | |
893 | * are only allowed for blocks use full backrefs. | |
894 | */ | |
895 | ||
896 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 897 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
898 | btrfs_header_level(buf), 1, |
899 | &refs, &flags); | |
be1a5564 MF |
900 | if (ret) |
901 | return ret; | |
e5df9573 MF |
902 | if (refs == 0) { |
903 | ret = -EROFS; | |
0b246afa | 904 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
905 | return ret; |
906 | } | |
5d4f98a2 YZ |
907 | } else { |
908 | refs = 1; | |
909 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
910 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
911 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
912 | else | |
913 | flags = 0; | |
914 | } | |
915 | ||
916 | owner = btrfs_header_owner(buf); | |
917 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
918 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
919 | ||
920 | if (refs > 1) { | |
921 | if ((owner == root->root_key.objectid || | |
922 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
923 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 924 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
925 | if (ret) |
926 | return ret; | |
5d4f98a2 YZ |
927 | |
928 | if (root->root_key.objectid == | |
929 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 930 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
931 | if (ret) |
932 | return ret; | |
e339a6b0 | 933 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
934 | if (ret) |
935 | return ret; | |
5d4f98a2 YZ |
936 | } |
937 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
938 | } else { | |
939 | ||
940 | if (root->root_key.objectid == | |
941 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 942 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 943 | else |
e339a6b0 | 944 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
945 | if (ret) |
946 | return ret; | |
5d4f98a2 YZ |
947 | } |
948 | if (new_flags != 0) { | |
b1c79e09 JB |
949 | int level = btrfs_header_level(buf); |
950 | ||
f5c8daa5 | 951 | ret = btrfs_set_disk_extent_flags(trans, |
5d4f98a2 YZ |
952 | buf->start, |
953 | buf->len, | |
b1c79e09 | 954 | new_flags, level, 0); |
be1a5564 MF |
955 | if (ret) |
956 | return ret; | |
5d4f98a2 YZ |
957 | } |
958 | } else { | |
959 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
960 | if (root->root_key.objectid == | |
961 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 962 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 963 | else |
e339a6b0 | 964 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
965 | if (ret) |
966 | return ret; | |
e339a6b0 | 967 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
968 | if (ret) |
969 | return ret; | |
5d4f98a2 | 970 | } |
6a884d7d | 971 | btrfs_clean_tree_block(buf); |
f0486c68 | 972 | *last_ref = 1; |
5d4f98a2 YZ |
973 | } |
974 | return 0; | |
975 | } | |
976 | ||
a6279470 FM |
977 | static struct extent_buffer *alloc_tree_block_no_bg_flush( |
978 | struct btrfs_trans_handle *trans, | |
979 | struct btrfs_root *root, | |
980 | u64 parent_start, | |
981 | const struct btrfs_disk_key *disk_key, | |
982 | int level, | |
983 | u64 hint, | |
984 | u64 empty_size) | |
985 | { | |
986 | struct btrfs_fs_info *fs_info = root->fs_info; | |
987 | struct extent_buffer *ret; | |
988 | ||
989 | /* | |
990 | * If we are COWing a node/leaf from the extent, chunk, device or free | |
991 | * space trees, make sure that we do not finish block group creation of | |
992 | * pending block groups. We do this to avoid a deadlock. | |
993 | * COWing can result in allocation of a new chunk, and flushing pending | |
994 | * block groups (btrfs_create_pending_block_groups()) can be triggered | |
995 | * when finishing allocation of a new chunk. Creation of a pending block | |
996 | * group modifies the extent, chunk, device and free space trees, | |
997 | * therefore we could deadlock with ourselves since we are holding a | |
998 | * lock on an extent buffer that btrfs_create_pending_block_groups() may | |
999 | * try to COW later. | |
1000 | * For similar reasons, we also need to delay flushing pending block | |
1001 | * groups when splitting a leaf or node, from one of those trees, since | |
1002 | * we are holding a write lock on it and its parent or when inserting a | |
1003 | * new root node for one of those trees. | |
1004 | */ | |
1005 | if (root == fs_info->extent_root || | |
1006 | root == fs_info->chunk_root || | |
1007 | root == fs_info->dev_root || | |
1008 | root == fs_info->free_space_root) | |
1009 | trans->can_flush_pending_bgs = false; | |
1010 | ||
1011 | ret = btrfs_alloc_tree_block(trans, root, parent_start, | |
1012 | root->root_key.objectid, disk_key, level, | |
1013 | hint, empty_size); | |
1014 | trans->can_flush_pending_bgs = true; | |
1015 | ||
1016 | return ret; | |
1017 | } | |
1018 | ||
d352ac68 | 1019 | /* |
d397712b CM |
1020 | * does the dirty work in cow of a single block. The parent block (if |
1021 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1022 | * dirty and returned locked. If you modify the block it needs to be marked | |
1023 | * dirty again. | |
d352ac68 CM |
1024 | * |
1025 | * search_start -- an allocation hint for the new block | |
1026 | * | |
d397712b CM |
1027 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1028 | * bytes the allocator should try to find free next to the block it returns. | |
1029 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1030 | */ |
d397712b | 1031 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1032 | struct btrfs_root *root, |
1033 | struct extent_buffer *buf, | |
1034 | struct extent_buffer *parent, int parent_slot, | |
1035 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1036 | u64 search_start, u64 empty_size) |
02217ed2 | 1037 | { |
0b246afa | 1038 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1039 | struct btrfs_disk_key disk_key; |
5f39d397 | 1040 | struct extent_buffer *cow; |
be1a5564 | 1041 | int level, ret; |
f0486c68 | 1042 | int last_ref = 0; |
925baedd | 1043 | int unlock_orig = 0; |
0f5053eb | 1044 | u64 parent_start = 0; |
7bb86316 | 1045 | |
925baedd CM |
1046 | if (*cow_ret == buf) |
1047 | unlock_orig = 1; | |
1048 | ||
b9447ef8 | 1049 | btrfs_assert_tree_locked(buf); |
925baedd | 1050 | |
27cdeb70 | 1051 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1052 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1053 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1054 | trans->transid != root->last_trans); | |
5f39d397 | 1055 | |
7bb86316 | 1056 | level = btrfs_header_level(buf); |
31840ae1 | 1057 | |
5d4f98a2 YZ |
1058 | if (level == 0) |
1059 | btrfs_item_key(buf, &disk_key, 0); | |
1060 | else | |
1061 | btrfs_node_key(buf, &disk_key, 0); | |
1062 | ||
0f5053eb GR |
1063 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1064 | parent_start = parent->start; | |
5d4f98a2 | 1065 | |
a6279470 FM |
1066 | cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key, |
1067 | level, search_start, empty_size); | |
54aa1f4d CM |
1068 | if (IS_ERR(cow)) |
1069 | return PTR_ERR(cow); | |
6702ed49 | 1070 | |
b4ce94de CM |
1071 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1072 | ||
58e8012c | 1073 | copy_extent_buffer_full(cow, buf); |
db94535d | 1074 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1075 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1076 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1077 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1078 | BTRFS_HEADER_FLAG_RELOC); | |
1079 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1080 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1081 | else | |
1082 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1083 | |
de37aa51 | 1084 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 1085 | |
be1a5564 | 1086 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1087 | if (ret) { |
66642832 | 1088 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1089 | return ret; |
1090 | } | |
1a40e23b | 1091 | |
27cdeb70 | 1092 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1093 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1094 | if (ret) { |
66642832 | 1095 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1096 | return ret; |
93314e3b | 1097 | } |
83d4cfd4 | 1098 | } |
3fd0a558 | 1099 | |
02217ed2 | 1100 | if (buf == root->node) { |
925baedd | 1101 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1102 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1103 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1104 | parent_start = buf->start; | |
925baedd | 1105 | |
5f39d397 | 1106 | extent_buffer_get(cow); |
d9d19a01 DS |
1107 | ret = tree_mod_log_insert_root(root->node, cow, 1); |
1108 | BUG_ON(ret < 0); | |
240f62c8 | 1109 | rcu_assign_pointer(root->node, cow); |
925baedd | 1110 | |
f0486c68 | 1111 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1112 | last_ref); |
5f39d397 | 1113 | free_extent_buffer(buf); |
0b86a832 | 1114 | add_root_to_dirty_list(root); |
02217ed2 | 1115 | } else { |
5d4f98a2 | 1116 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
e09c2efe | 1117 | tree_mod_log_insert_key(parent, parent_slot, |
c8cc6341 | 1118 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1119 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1120 | cow->start); |
74493f7a CM |
1121 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1122 | trans->transid); | |
d6025579 | 1123 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1124 | if (last_ref) { |
db7279a2 | 1125 | ret = tree_mod_log_free_eb(buf); |
5de865ee | 1126 | if (ret) { |
66642832 | 1127 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1128 | return ret; |
1129 | } | |
1130 | } | |
f0486c68 | 1131 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1132 | last_ref); |
02217ed2 | 1133 | } |
925baedd CM |
1134 | if (unlock_orig) |
1135 | btrfs_tree_unlock(buf); | |
3083ee2e | 1136 | free_extent_buffer_stale(buf); |
ccd467d6 | 1137 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1138 | *cow_ret = cow; |
02217ed2 CM |
1139 | return 0; |
1140 | } | |
1141 | ||
5d9e75c4 JS |
1142 | /* |
1143 | * returns the logical address of the oldest predecessor of the given root. | |
1144 | * entries older than time_seq are ignored. | |
1145 | */ | |
bcd24dab DS |
1146 | static struct tree_mod_elem *__tree_mod_log_oldest_root( |
1147 | struct extent_buffer *eb_root, u64 time_seq) | |
5d9e75c4 JS |
1148 | { |
1149 | struct tree_mod_elem *tm; | |
1150 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1151 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1152 | int looped = 0; |
1153 | ||
1154 | if (!time_seq) | |
35a3621b | 1155 | return NULL; |
5d9e75c4 JS |
1156 | |
1157 | /* | |
298cfd36 CR |
1158 | * the very last operation that's logged for a root is the |
1159 | * replacement operation (if it is replaced at all). this has | |
1160 | * the logical address of the *new* root, making it the very | |
1161 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1162 | */ |
1163 | while (1) { | |
bcd24dab | 1164 | tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical, |
5d9e75c4 JS |
1165 | time_seq); |
1166 | if (!looped && !tm) | |
35a3621b | 1167 | return NULL; |
5d9e75c4 | 1168 | /* |
28da9fb4 JS |
1169 | * if there are no tree operation for the oldest root, we simply |
1170 | * return it. this should only happen if that (old) root is at | |
1171 | * level 0. | |
5d9e75c4 | 1172 | */ |
28da9fb4 JS |
1173 | if (!tm) |
1174 | break; | |
5d9e75c4 | 1175 | |
28da9fb4 JS |
1176 | /* |
1177 | * if there's an operation that's not a root replacement, we | |
1178 | * found the oldest version of our root. normally, we'll find a | |
1179 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1180 | */ | |
5d9e75c4 JS |
1181 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1182 | break; | |
1183 | ||
1184 | found = tm; | |
1185 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1186 | looped = 1; |
1187 | } | |
1188 | ||
a95236d9 JS |
1189 | /* if there's no old root to return, return what we found instead */ |
1190 | if (!found) | |
1191 | found = tm; | |
1192 | ||
5d9e75c4 JS |
1193 | return found; |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1198 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1199 | * time_seq). |
1200 | */ | |
1201 | static void | |
f1ca7e98 JB |
1202 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1203 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1204 | { |
1205 | u32 n; | |
1206 | struct rb_node *next; | |
1207 | struct tree_mod_elem *tm = first_tm; | |
1208 | unsigned long o_dst; | |
1209 | unsigned long o_src; | |
1210 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1211 | ||
1212 | n = btrfs_header_nritems(eb); | |
b1a09f1e | 1213 | read_lock(&fs_info->tree_mod_log_lock); |
097b8a7c | 1214 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1215 | /* |
1216 | * all the operations are recorded with the operator used for | |
1217 | * the modification. as we're going backwards, we do the | |
1218 | * opposite of each operation here. | |
1219 | */ | |
1220 | switch (tm->op) { | |
1221 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1222 | BUG_ON(tm->slot < n); | |
1c697d4a | 1223 | /* Fallthrough */ |
95c80bb1 | 1224 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1225 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1226 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1227 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1228 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1229 | tm->generation); | |
4c3e6969 | 1230 | n++; |
5d9e75c4 JS |
1231 | break; |
1232 | case MOD_LOG_KEY_REPLACE: | |
1233 | BUG_ON(tm->slot >= n); | |
1234 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1235 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1236 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1237 | tm->generation); | |
1238 | break; | |
1239 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1240 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1241 | n--; |
1242 | break; | |
1243 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1244 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1245 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1246 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1247 | tm->move.nr_items * p_size); |
1248 | break; | |
1249 | case MOD_LOG_ROOT_REPLACE: | |
1250 | /* | |
1251 | * this operation is special. for roots, this must be | |
1252 | * handled explicitly before rewinding. | |
1253 | * for non-roots, this operation may exist if the node | |
1254 | * was a root: root A -> child B; then A gets empty and | |
1255 | * B is promoted to the new root. in the mod log, we'll | |
1256 | * have a root-replace operation for B, a tree block | |
1257 | * that is no root. we simply ignore that operation. | |
1258 | */ | |
1259 | break; | |
1260 | } | |
1261 | next = rb_next(&tm->node); | |
1262 | if (!next) | |
1263 | break; | |
6b4df8b6 | 1264 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1265 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1266 | break; |
1267 | } | |
b1a09f1e | 1268 | read_unlock(&fs_info->tree_mod_log_lock); |
5d9e75c4 JS |
1269 | btrfs_set_header_nritems(eb, n); |
1270 | } | |
1271 | ||
47fb091f | 1272 | /* |
01327610 | 1273 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1274 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1275 | * returned buffer is always read-locked. If the returned buffer is not the | |
1276 | * input buffer, the lock on the input buffer is released and the input buffer | |
1277 | * is freed (its refcount is decremented). | |
1278 | */ | |
5d9e75c4 | 1279 | static struct extent_buffer * |
9ec72677 JB |
1280 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1281 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1282 | { |
1283 | struct extent_buffer *eb_rewin; | |
1284 | struct tree_mod_elem *tm; | |
1285 | ||
1286 | if (!time_seq) | |
1287 | return eb; | |
1288 | ||
1289 | if (btrfs_header_level(eb) == 0) | |
1290 | return eb; | |
1291 | ||
1292 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1293 | if (!tm) | |
1294 | return eb; | |
1295 | ||
9ec72677 | 1296 | btrfs_set_path_blocking(path); |
300aa896 | 1297 | btrfs_set_lock_blocking_read(eb); |
9ec72677 | 1298 | |
5d9e75c4 JS |
1299 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1300 | BUG_ON(tm->slot != 0); | |
da17066c | 1301 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1302 | if (!eb_rewin) { |
9ec72677 | 1303 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1304 | free_extent_buffer(eb); |
1305 | return NULL; | |
1306 | } | |
5d9e75c4 JS |
1307 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1308 | btrfs_set_header_backref_rev(eb_rewin, | |
1309 | btrfs_header_backref_rev(eb)); | |
1310 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1311 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1312 | } else { |
1313 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1314 | if (!eb_rewin) { |
9ec72677 | 1315 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1316 | free_extent_buffer(eb); |
1317 | return NULL; | |
1318 | } | |
5d9e75c4 JS |
1319 | } |
1320 | ||
9ec72677 | 1321 | btrfs_tree_read_unlock_blocking(eb); |
5d9e75c4 JS |
1322 | free_extent_buffer(eb); |
1323 | ||
47fb091f | 1324 | btrfs_tree_read_lock(eb_rewin); |
f1ca7e98 | 1325 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1326 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1327 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1328 | |
1329 | return eb_rewin; | |
1330 | } | |
1331 | ||
8ba97a15 JS |
1332 | /* |
1333 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1334 | * value. If there are no changes, the current root->root_node is returned. If | |
1335 | * anything changed in between, there's a fresh buffer allocated on which the | |
1336 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1337 | * Returns NULL on error (with no locks held). | |
1338 | */ | |
5d9e75c4 JS |
1339 | static inline struct extent_buffer * |
1340 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1341 | { | |
0b246afa | 1342 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1343 | struct tree_mod_elem *tm; |
30b0463a JS |
1344 | struct extent_buffer *eb = NULL; |
1345 | struct extent_buffer *eb_root; | |
efad8a85 | 1346 | u64 eb_root_owner = 0; |
7bfdcf7f | 1347 | struct extent_buffer *old; |
a95236d9 | 1348 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1349 | u64 old_generation = 0; |
a95236d9 | 1350 | u64 logical; |
581c1760 | 1351 | int level; |
5d9e75c4 | 1352 | |
30b0463a | 1353 | eb_root = btrfs_read_lock_root_node(root); |
bcd24dab | 1354 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5d9e75c4 | 1355 | if (!tm) |
30b0463a | 1356 | return eb_root; |
5d9e75c4 | 1357 | |
a95236d9 JS |
1358 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1359 | old_root = &tm->old_root; | |
1360 | old_generation = tm->generation; | |
1361 | logical = old_root->logical; | |
581c1760 | 1362 | level = old_root->level; |
a95236d9 | 1363 | } else { |
30b0463a | 1364 | logical = eb_root->start; |
581c1760 | 1365 | level = btrfs_header_level(eb_root); |
a95236d9 | 1366 | } |
5d9e75c4 | 1367 | |
0b246afa | 1368 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1369 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1370 | btrfs_tree_read_unlock(eb_root); |
1371 | free_extent_buffer(eb_root); | |
581c1760 | 1372 | old = read_tree_block(fs_info, logical, 0, level, NULL); |
64c043de LB |
1373 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1374 | if (!IS_ERR(old)) | |
1375 | free_extent_buffer(old); | |
0b246afa JM |
1376 | btrfs_warn(fs_info, |
1377 | "failed to read tree block %llu from get_old_root", | |
1378 | logical); | |
834328a8 | 1379 | } else { |
7bfdcf7f LB |
1380 | eb = btrfs_clone_extent_buffer(old); |
1381 | free_extent_buffer(old); | |
834328a8 JS |
1382 | } |
1383 | } else if (old_root) { | |
efad8a85 | 1384 | eb_root_owner = btrfs_header_owner(eb_root); |
30b0463a JS |
1385 | btrfs_tree_read_unlock(eb_root); |
1386 | free_extent_buffer(eb_root); | |
0b246afa | 1387 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1388 | } else { |
300aa896 | 1389 | btrfs_set_lock_blocking_read(eb_root); |
30b0463a | 1390 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1391 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1392 | free_extent_buffer(eb_root); |
834328a8 JS |
1393 | } |
1394 | ||
8ba97a15 JS |
1395 | if (!eb) |
1396 | return NULL; | |
1397 | btrfs_tree_read_lock(eb); | |
a95236d9 | 1398 | if (old_root) { |
5d9e75c4 JS |
1399 | btrfs_set_header_bytenr(eb, eb->start); |
1400 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
efad8a85 | 1401 | btrfs_set_header_owner(eb, eb_root_owner); |
a95236d9 JS |
1402 | btrfs_set_header_level(eb, old_root->level); |
1403 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1404 | } |
28da9fb4 | 1405 | if (tm) |
0b246afa | 1406 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1407 | else |
1408 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1409 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1410 | |
1411 | return eb; | |
1412 | } | |
1413 | ||
5b6602e7 JS |
1414 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1415 | { | |
1416 | struct tree_mod_elem *tm; | |
1417 | int level; | |
30b0463a | 1418 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1419 | |
bcd24dab | 1420 | tm = __tree_mod_log_oldest_root(eb_root, time_seq); |
5b6602e7 JS |
1421 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1422 | level = tm->old_root.level; | |
1423 | } else { | |
30b0463a | 1424 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1425 | } |
30b0463a | 1426 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1427 | |
1428 | return level; | |
1429 | } | |
1430 | ||
5d4f98a2 YZ |
1431 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1432 | struct btrfs_root *root, | |
1433 | struct extent_buffer *buf) | |
1434 | { | |
f5ee5c9a | 1435 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1436 | return 0; |
fccb84c9 | 1437 | |
d1980131 DS |
1438 | /* Ensure we can see the FORCE_COW bit */ |
1439 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
1440 | |
1441 | /* | |
1442 | * We do not need to cow a block if | |
1443 | * 1) this block is not created or changed in this transaction; | |
1444 | * 2) this block does not belong to TREE_RELOC tree; | |
1445 | * 3) the root is not forced COW. | |
1446 | * | |
1447 | * What is forced COW: | |
01327610 | 1448 | * when we create snapshot during committing the transaction, |
52042d8e | 1449 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
1450 | * block to ensure the metadata consistency. |
1451 | */ | |
5d4f98a2 YZ |
1452 | if (btrfs_header_generation(buf) == trans->transid && |
1453 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1454 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1455 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1456 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1457 | return 0; |
1458 | return 1; | |
1459 | } | |
1460 | ||
d352ac68 CM |
1461 | /* |
1462 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1463 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1464 | * once per transaction, as long as it hasn't been written yet |
1465 | */ | |
d397712b | 1466 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1467 | struct btrfs_root *root, struct extent_buffer *buf, |
1468 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1469 | struct extent_buffer **cow_ret) |
6702ed49 | 1470 | { |
0b246afa | 1471 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1472 | u64 search_start; |
f510cfec | 1473 | int ret; |
dc17ff8f | 1474 | |
83354f07 JB |
1475 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
1476 | btrfs_err(fs_info, | |
1477 | "COW'ing blocks on a fs root that's being dropped"); | |
1478 | ||
0b246afa | 1479 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1480 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1481 | trans->transid, |
0b246afa | 1482 | fs_info->running_transaction->transid); |
31b1a2bd | 1483 | |
0b246afa | 1484 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1485 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1486 | trans->transid, fs_info->generation); |
dc17ff8f | 1487 | |
5d4f98a2 | 1488 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1489 | trans->dirty = true; |
6702ed49 CM |
1490 | *cow_ret = buf; |
1491 | return 0; | |
1492 | } | |
c487685d | 1493 | |
ee22184b | 1494 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1495 | |
1496 | if (parent) | |
8bead258 DS |
1497 | btrfs_set_lock_blocking_write(parent); |
1498 | btrfs_set_lock_blocking_write(buf); | |
b4ce94de | 1499 | |
f616f5cd QW |
1500 | /* |
1501 | * Before CoWing this block for later modification, check if it's | |
1502 | * the subtree root and do the delayed subtree trace if needed. | |
1503 | * | |
1504 | * Also We don't care about the error, as it's handled internally. | |
1505 | */ | |
1506 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 1507 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1508 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1509 | |
1510 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1511 | ||
f510cfec | 1512 | return ret; |
6702ed49 CM |
1513 | } |
1514 | ||
d352ac68 CM |
1515 | /* |
1516 | * helper function for defrag to decide if two blocks pointed to by a | |
1517 | * node are actually close by | |
1518 | */ | |
6b80053d | 1519 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1520 | { |
6b80053d | 1521 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1522 | return 1; |
6b80053d | 1523 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1524 | return 1; |
1525 | return 0; | |
1526 | } | |
1527 | ||
081e9573 CM |
1528 | /* |
1529 | * compare two keys in a memcmp fashion | |
1530 | */ | |
310712b2 OS |
1531 | static int comp_keys(const struct btrfs_disk_key *disk, |
1532 | const struct btrfs_key *k2) | |
081e9573 CM |
1533 | { |
1534 | struct btrfs_key k1; | |
1535 | ||
1536 | btrfs_disk_key_to_cpu(&k1, disk); | |
1537 | ||
20736aba | 1538 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1539 | } |
1540 | ||
f3465ca4 JB |
1541 | /* |
1542 | * same as comp_keys only with two btrfs_key's | |
1543 | */ | |
310712b2 | 1544 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
1545 | { |
1546 | if (k1->objectid > k2->objectid) | |
1547 | return 1; | |
1548 | if (k1->objectid < k2->objectid) | |
1549 | return -1; | |
1550 | if (k1->type > k2->type) | |
1551 | return 1; | |
1552 | if (k1->type < k2->type) | |
1553 | return -1; | |
1554 | if (k1->offset > k2->offset) | |
1555 | return 1; | |
1556 | if (k1->offset < k2->offset) | |
1557 | return -1; | |
1558 | return 0; | |
1559 | } | |
081e9573 | 1560 | |
d352ac68 CM |
1561 | /* |
1562 | * this is used by the defrag code to go through all the | |
1563 | * leaves pointed to by a node and reallocate them so that | |
1564 | * disk order is close to key order | |
1565 | */ | |
6702ed49 | 1566 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1567 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1568 | int start_slot, u64 *last_ret, |
a6b6e75e | 1569 | struct btrfs_key *progress) |
6702ed49 | 1570 | { |
0b246afa | 1571 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1572 | struct extent_buffer *cur; |
6702ed49 | 1573 | u64 blocknr; |
ca7a79ad | 1574 | u64 gen; |
e9d0b13b CM |
1575 | u64 search_start = *last_ret; |
1576 | u64 last_block = 0; | |
6702ed49 CM |
1577 | u64 other; |
1578 | u32 parent_nritems; | |
6702ed49 CM |
1579 | int end_slot; |
1580 | int i; | |
1581 | int err = 0; | |
f2183bde | 1582 | int parent_level; |
6b80053d CM |
1583 | int uptodate; |
1584 | u32 blocksize; | |
081e9573 CM |
1585 | int progress_passed = 0; |
1586 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1587 | |
5708b959 | 1588 | parent_level = btrfs_header_level(parent); |
5708b959 | 1589 | |
0b246afa JM |
1590 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1591 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1592 | |
6b80053d | 1593 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1594 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1595 | end_slot = parent_nritems - 1; |
6702ed49 | 1596 | |
5dfe2be7 | 1597 | if (parent_nritems <= 1) |
6702ed49 CM |
1598 | return 0; |
1599 | ||
8bead258 | 1600 | btrfs_set_lock_blocking_write(parent); |
b4ce94de | 1601 | |
5dfe2be7 | 1602 | for (i = start_slot; i <= end_slot; i++) { |
581c1760 | 1603 | struct btrfs_key first_key; |
6702ed49 | 1604 | int close = 1; |
a6b6e75e | 1605 | |
081e9573 CM |
1606 | btrfs_node_key(parent, &disk_key, i); |
1607 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1608 | continue; | |
1609 | ||
1610 | progress_passed = 1; | |
6b80053d | 1611 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1612 | gen = btrfs_node_ptr_generation(parent, i); |
581c1760 | 1613 | btrfs_node_key_to_cpu(parent, &first_key, i); |
e9d0b13b CM |
1614 | if (last_block == 0) |
1615 | last_block = blocknr; | |
5708b959 | 1616 | |
6702ed49 | 1617 | if (i > 0) { |
6b80053d CM |
1618 | other = btrfs_node_blockptr(parent, i - 1); |
1619 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1620 | } |
5dfe2be7 | 1621 | if (!close && i < end_slot) { |
6b80053d CM |
1622 | other = btrfs_node_blockptr(parent, i + 1); |
1623 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1624 | } |
e9d0b13b CM |
1625 | if (close) { |
1626 | last_block = blocknr; | |
6702ed49 | 1627 | continue; |
e9d0b13b | 1628 | } |
6702ed49 | 1629 | |
0b246afa | 1630 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1631 | if (cur) |
b9fab919 | 1632 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1633 | else |
1634 | uptodate = 0; | |
5708b959 | 1635 | if (!cur || !uptodate) { |
6b80053d | 1636 | if (!cur) { |
581c1760 QW |
1637 | cur = read_tree_block(fs_info, blocknr, gen, |
1638 | parent_level - 1, | |
1639 | &first_key); | |
64c043de LB |
1640 | if (IS_ERR(cur)) { |
1641 | return PTR_ERR(cur); | |
1642 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1643 | free_extent_buffer(cur); |
97d9a8a4 | 1644 | return -EIO; |
416bc658 | 1645 | } |
6b80053d | 1646 | } else if (!uptodate) { |
581c1760 QW |
1647 | err = btrfs_read_buffer(cur, gen, |
1648 | parent_level - 1,&first_key); | |
018642a1 TI |
1649 | if (err) { |
1650 | free_extent_buffer(cur); | |
1651 | return err; | |
1652 | } | |
f2183bde | 1653 | } |
6702ed49 | 1654 | } |
e9d0b13b | 1655 | if (search_start == 0) |
6b80053d | 1656 | search_start = last_block; |
e9d0b13b | 1657 | |
e7a84565 | 1658 | btrfs_tree_lock(cur); |
8bead258 | 1659 | btrfs_set_lock_blocking_write(cur); |
6b80053d | 1660 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1661 | &cur, search_start, |
6b80053d | 1662 | min(16 * blocksize, |
9fa8cfe7 | 1663 | (end_slot - i) * blocksize)); |
252c38f0 | 1664 | if (err) { |
e7a84565 | 1665 | btrfs_tree_unlock(cur); |
6b80053d | 1666 | free_extent_buffer(cur); |
6702ed49 | 1667 | break; |
252c38f0 | 1668 | } |
e7a84565 CM |
1669 | search_start = cur->start; |
1670 | last_block = cur->start; | |
f2183bde | 1671 | *last_ret = search_start; |
e7a84565 CM |
1672 | btrfs_tree_unlock(cur); |
1673 | free_extent_buffer(cur); | |
6702ed49 CM |
1674 | } |
1675 | return err; | |
1676 | } | |
1677 | ||
74123bd7 | 1678 | /* |
5f39d397 CM |
1679 | * search for key in the extent_buffer. The items start at offset p, |
1680 | * and they are item_size apart. There are 'max' items in p. | |
1681 | * | |
74123bd7 CM |
1682 | * the slot in the array is returned via slot, and it points to |
1683 | * the place where you would insert key if it is not found in | |
1684 | * the array. | |
1685 | * | |
1686 | * slot may point to max if the key is bigger than all of the keys | |
1687 | */ | |
e02119d5 | 1688 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1689 | unsigned long p, int item_size, |
1690 | const struct btrfs_key *key, | |
e02119d5 | 1691 | int max, int *slot) |
be0e5c09 CM |
1692 | { |
1693 | int low = 0; | |
1694 | int high = max; | |
1695 | int mid; | |
1696 | int ret; | |
479965d6 | 1697 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1698 | struct btrfs_disk_key unaligned; |
1699 | unsigned long offset; | |
5f39d397 CM |
1700 | char *kaddr = NULL; |
1701 | unsigned long map_start = 0; | |
1702 | unsigned long map_len = 0; | |
479965d6 | 1703 | int err; |
be0e5c09 | 1704 | |
5e24e9af LB |
1705 | if (low > high) { |
1706 | btrfs_err(eb->fs_info, | |
1707 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1708 | __func__, low, high, eb->start, | |
1709 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1710 | return -EINVAL; | |
1711 | } | |
1712 | ||
d397712b | 1713 | while (low < high) { |
be0e5c09 | 1714 | mid = (low + high) / 2; |
5f39d397 CM |
1715 | offset = p + mid * item_size; |
1716 | ||
a6591715 | 1717 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1718 | (offset + sizeof(struct btrfs_disk_key)) > |
1719 | map_start + map_len) { | |
934d375b CM |
1720 | |
1721 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1722 | sizeof(struct btrfs_disk_key), |
a6591715 | 1723 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1724 | |
1725 | if (!err) { | |
1726 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1727 | map_start); | |
415b35a5 | 1728 | } else if (err == 1) { |
479965d6 CM |
1729 | read_extent_buffer(eb, &unaligned, |
1730 | offset, sizeof(unaligned)); | |
1731 | tmp = &unaligned; | |
415b35a5 LB |
1732 | } else { |
1733 | return err; | |
479965d6 | 1734 | } |
5f39d397 | 1735 | |
5f39d397 CM |
1736 | } else { |
1737 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1738 | map_start); | |
1739 | } | |
be0e5c09 CM |
1740 | ret = comp_keys(tmp, key); |
1741 | ||
1742 | if (ret < 0) | |
1743 | low = mid + 1; | |
1744 | else if (ret > 0) | |
1745 | high = mid; | |
1746 | else { | |
1747 | *slot = mid; | |
1748 | return 0; | |
1749 | } | |
1750 | } | |
1751 | *slot = low; | |
1752 | return 1; | |
1753 | } | |
1754 | ||
97571fd0 CM |
1755 | /* |
1756 | * simple bin_search frontend that does the right thing for | |
1757 | * leaves vs nodes | |
1758 | */ | |
a74b35ec NB |
1759 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
1760 | int level, int *slot) | |
be0e5c09 | 1761 | { |
f775738f | 1762 | if (level == 0) |
5f39d397 CM |
1763 | return generic_bin_search(eb, |
1764 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1765 | sizeof(struct btrfs_item), |
5f39d397 | 1766 | key, btrfs_header_nritems(eb), |
7518a238 | 1767 | slot); |
f775738f | 1768 | else |
5f39d397 CM |
1769 | return generic_bin_search(eb, |
1770 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1771 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1772 | key, btrfs_header_nritems(eb), |
7518a238 | 1773 | slot); |
be0e5c09 CM |
1774 | } |
1775 | ||
f0486c68 YZ |
1776 | static void root_add_used(struct btrfs_root *root, u32 size) |
1777 | { | |
1778 | spin_lock(&root->accounting_lock); | |
1779 | btrfs_set_root_used(&root->root_item, | |
1780 | btrfs_root_used(&root->root_item) + size); | |
1781 | spin_unlock(&root->accounting_lock); | |
1782 | } | |
1783 | ||
1784 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1785 | { | |
1786 | spin_lock(&root->accounting_lock); | |
1787 | btrfs_set_root_used(&root->root_item, | |
1788 | btrfs_root_used(&root->root_item) - size); | |
1789 | spin_unlock(&root->accounting_lock); | |
1790 | } | |
1791 | ||
d352ac68 CM |
1792 | /* given a node and slot number, this reads the blocks it points to. The |
1793 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1794 | */ |
4b231ae4 DS |
1795 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
1796 | int slot) | |
bb803951 | 1797 | { |
ca7a79ad | 1798 | int level = btrfs_header_level(parent); |
416bc658 | 1799 | struct extent_buffer *eb; |
581c1760 | 1800 | struct btrfs_key first_key; |
416bc658 | 1801 | |
fb770ae4 LB |
1802 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1803 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1804 | |
1805 | BUG_ON(level == 0); | |
1806 | ||
581c1760 | 1807 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 1808 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
581c1760 QW |
1809 | btrfs_node_ptr_generation(parent, slot), |
1810 | level - 1, &first_key); | |
fb770ae4 LB |
1811 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1812 | free_extent_buffer(eb); | |
1813 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1814 | } |
1815 | ||
1816 | return eb; | |
bb803951 CM |
1817 | } |
1818 | ||
d352ac68 CM |
1819 | /* |
1820 | * node level balancing, used to make sure nodes are in proper order for | |
1821 | * item deletion. We balance from the top down, so we have to make sure | |
1822 | * that a deletion won't leave an node completely empty later on. | |
1823 | */ | |
e02119d5 | 1824 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1825 | struct btrfs_root *root, |
1826 | struct btrfs_path *path, int level) | |
bb803951 | 1827 | { |
0b246afa | 1828 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1829 | struct extent_buffer *right = NULL; |
1830 | struct extent_buffer *mid; | |
1831 | struct extent_buffer *left = NULL; | |
1832 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1833 | int ret = 0; |
1834 | int wret; | |
1835 | int pslot; | |
bb803951 | 1836 | int orig_slot = path->slots[level]; |
79f95c82 | 1837 | u64 orig_ptr; |
bb803951 | 1838 | |
98e6b1eb | 1839 | ASSERT(level > 0); |
bb803951 | 1840 | |
5f39d397 | 1841 | mid = path->nodes[level]; |
b4ce94de | 1842 | |
bd681513 CM |
1843 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1844 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1845 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1846 | ||
1d4f8a0c | 1847 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1848 | |
a05a9bb1 | 1849 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1850 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1851 | pslot = path->slots[level + 1]; |
1852 | } | |
bb803951 | 1853 | |
40689478 CM |
1854 | /* |
1855 | * deal with the case where there is only one pointer in the root | |
1856 | * by promoting the node below to a root | |
1857 | */ | |
5f39d397 CM |
1858 | if (!parent) { |
1859 | struct extent_buffer *child; | |
bb803951 | 1860 | |
5f39d397 | 1861 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1862 | return 0; |
1863 | ||
1864 | /* promote the child to a root */ | |
4b231ae4 | 1865 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
1866 | if (IS_ERR(child)) { |
1867 | ret = PTR_ERR(child); | |
0b246afa | 1868 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1869 | goto enospc; |
1870 | } | |
1871 | ||
925baedd | 1872 | btrfs_tree_lock(child); |
8bead258 | 1873 | btrfs_set_lock_blocking_write(child); |
9fa8cfe7 | 1874 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1875 | if (ret) { |
1876 | btrfs_tree_unlock(child); | |
1877 | free_extent_buffer(child); | |
1878 | goto enospc; | |
1879 | } | |
2f375ab9 | 1880 | |
d9d19a01 DS |
1881 | ret = tree_mod_log_insert_root(root->node, child, 1); |
1882 | BUG_ON(ret < 0); | |
240f62c8 | 1883 | rcu_assign_pointer(root->node, child); |
925baedd | 1884 | |
0b86a832 | 1885 | add_root_to_dirty_list(root); |
925baedd | 1886 | btrfs_tree_unlock(child); |
b4ce94de | 1887 | |
925baedd | 1888 | path->locks[level] = 0; |
bb803951 | 1889 | path->nodes[level] = NULL; |
6a884d7d | 1890 | btrfs_clean_tree_block(mid); |
925baedd | 1891 | btrfs_tree_unlock(mid); |
bb803951 | 1892 | /* once for the path */ |
5f39d397 | 1893 | free_extent_buffer(mid); |
f0486c68 YZ |
1894 | |
1895 | root_sub_used(root, mid->len); | |
5581a51a | 1896 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1897 | /* once for the root ptr */ |
3083ee2e | 1898 | free_extent_buffer_stale(mid); |
f0486c68 | 1899 | return 0; |
bb803951 | 1900 | } |
5f39d397 | 1901 | if (btrfs_header_nritems(mid) > |
0b246afa | 1902 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1903 | return 0; |
1904 | ||
4b231ae4 | 1905 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1906 | if (IS_ERR(left)) |
1907 | left = NULL; | |
1908 | ||
5f39d397 | 1909 | if (left) { |
925baedd | 1910 | btrfs_tree_lock(left); |
8bead258 | 1911 | btrfs_set_lock_blocking_write(left); |
5f39d397 | 1912 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1913 | parent, pslot - 1, &left); |
54aa1f4d CM |
1914 | if (wret) { |
1915 | ret = wret; | |
1916 | goto enospc; | |
1917 | } | |
2cc58cf2 | 1918 | } |
fb770ae4 | 1919 | |
4b231ae4 | 1920 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1921 | if (IS_ERR(right)) |
1922 | right = NULL; | |
1923 | ||
5f39d397 | 1924 | if (right) { |
925baedd | 1925 | btrfs_tree_lock(right); |
8bead258 | 1926 | btrfs_set_lock_blocking_write(right); |
5f39d397 | 1927 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1928 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1929 | if (wret) { |
1930 | ret = wret; | |
1931 | goto enospc; | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1936 | if (left) { |
1937 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1938 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1939 | if (wret < 0) |
1940 | ret = wret; | |
bb803951 | 1941 | } |
79f95c82 CM |
1942 | |
1943 | /* | |
1944 | * then try to empty the right most buffer into the middle | |
1945 | */ | |
5f39d397 | 1946 | if (right) { |
d30a668f | 1947 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1948 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1949 | ret = wret; |
5f39d397 | 1950 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1951 | btrfs_clean_tree_block(right); |
925baedd | 1952 | btrfs_tree_unlock(right); |
afe5fea7 | 1953 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1954 | root_sub_used(root, right->len); |
5581a51a | 1955 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1956 | free_extent_buffer_stale(right); |
f0486c68 | 1957 | right = NULL; |
bb803951 | 1958 | } else { |
5f39d397 CM |
1959 | struct btrfs_disk_key right_key; |
1960 | btrfs_node_key(right, &right_key, 0); | |
0e82bcfe DS |
1961 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
1962 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
1963 | BUG_ON(ret < 0); | |
5f39d397 CM |
1964 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1965 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1966 | } |
1967 | } | |
5f39d397 | 1968 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1969 | /* |
1970 | * we're not allowed to leave a node with one item in the | |
1971 | * tree during a delete. A deletion from lower in the tree | |
1972 | * could try to delete the only pointer in this node. | |
1973 | * So, pull some keys from the left. | |
1974 | * There has to be a left pointer at this point because | |
1975 | * otherwise we would have pulled some pointers from the | |
1976 | * right | |
1977 | */ | |
305a26af MF |
1978 | if (!left) { |
1979 | ret = -EROFS; | |
0b246afa | 1980 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1981 | goto enospc; |
1982 | } | |
55d32ed8 | 1983 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1984 | if (wret < 0) { |
79f95c82 | 1985 | ret = wret; |
54aa1f4d CM |
1986 | goto enospc; |
1987 | } | |
bce4eae9 | 1988 | if (wret == 1) { |
d30a668f | 1989 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1990 | if (wret < 0) |
1991 | ret = wret; | |
1992 | } | |
79f95c82 CM |
1993 | BUG_ON(wret == 1); |
1994 | } | |
5f39d397 | 1995 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1996 | btrfs_clean_tree_block(mid); |
925baedd | 1997 | btrfs_tree_unlock(mid); |
afe5fea7 | 1998 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1999 | root_sub_used(root, mid->len); |
5581a51a | 2000 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2001 | free_extent_buffer_stale(mid); |
f0486c68 | 2002 | mid = NULL; |
79f95c82 CM |
2003 | } else { |
2004 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2005 | struct btrfs_disk_key mid_key; |
2006 | btrfs_node_key(mid, &mid_key, 0); | |
0e82bcfe DS |
2007 | ret = tree_mod_log_insert_key(parent, pslot, |
2008 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2009 | BUG_ON(ret < 0); | |
5f39d397 CM |
2010 | btrfs_set_node_key(parent, &mid_key, pslot); |
2011 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2012 | } |
bb803951 | 2013 | |
79f95c82 | 2014 | /* update the path */ |
5f39d397 CM |
2015 | if (left) { |
2016 | if (btrfs_header_nritems(left) > orig_slot) { | |
2017 | extent_buffer_get(left); | |
925baedd | 2018 | /* left was locked after cow */ |
5f39d397 | 2019 | path->nodes[level] = left; |
bb803951 CM |
2020 | path->slots[level + 1] -= 1; |
2021 | path->slots[level] = orig_slot; | |
925baedd CM |
2022 | if (mid) { |
2023 | btrfs_tree_unlock(mid); | |
5f39d397 | 2024 | free_extent_buffer(mid); |
925baedd | 2025 | } |
bb803951 | 2026 | } else { |
5f39d397 | 2027 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2028 | path->slots[level] = orig_slot; |
2029 | } | |
2030 | } | |
79f95c82 | 2031 | /* double check we haven't messed things up */ |
e20d96d6 | 2032 | if (orig_ptr != |
5f39d397 | 2033 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2034 | BUG(); |
54aa1f4d | 2035 | enospc: |
925baedd CM |
2036 | if (right) { |
2037 | btrfs_tree_unlock(right); | |
5f39d397 | 2038 | free_extent_buffer(right); |
925baedd CM |
2039 | } |
2040 | if (left) { | |
2041 | if (path->nodes[level] != left) | |
2042 | btrfs_tree_unlock(left); | |
5f39d397 | 2043 | free_extent_buffer(left); |
925baedd | 2044 | } |
bb803951 CM |
2045 | return ret; |
2046 | } | |
2047 | ||
d352ac68 CM |
2048 | /* Node balancing for insertion. Here we only split or push nodes around |
2049 | * when they are completely full. This is also done top down, so we | |
2050 | * have to be pessimistic. | |
2051 | */ | |
d397712b | 2052 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2053 | struct btrfs_root *root, |
2054 | struct btrfs_path *path, int level) | |
e66f709b | 2055 | { |
0b246afa | 2056 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2057 | struct extent_buffer *right = NULL; |
2058 | struct extent_buffer *mid; | |
2059 | struct extent_buffer *left = NULL; | |
2060 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2061 | int ret = 0; |
2062 | int wret; | |
2063 | int pslot; | |
2064 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2065 | |
2066 | if (level == 0) | |
2067 | return 1; | |
2068 | ||
5f39d397 | 2069 | mid = path->nodes[level]; |
7bb86316 | 2070 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2071 | |
a05a9bb1 | 2072 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2073 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2074 | pslot = path->slots[level + 1]; |
2075 | } | |
e66f709b | 2076 | |
5f39d397 | 2077 | if (!parent) |
e66f709b | 2078 | return 1; |
e66f709b | 2079 | |
4b231ae4 | 2080 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
2081 | if (IS_ERR(left)) |
2082 | left = NULL; | |
e66f709b CM |
2083 | |
2084 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2085 | if (left) { |
e66f709b | 2086 | u32 left_nr; |
925baedd CM |
2087 | |
2088 | btrfs_tree_lock(left); | |
8bead258 | 2089 | btrfs_set_lock_blocking_write(left); |
b4ce94de | 2090 | |
5f39d397 | 2091 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2092 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2093 | wret = 1; |
2094 | } else { | |
5f39d397 | 2095 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2096 | pslot - 1, &left); |
54aa1f4d CM |
2097 | if (ret) |
2098 | wret = 1; | |
2099 | else { | |
d30a668f | 2100 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 2101 | } |
33ade1f8 | 2102 | } |
e66f709b CM |
2103 | if (wret < 0) |
2104 | ret = wret; | |
2105 | if (wret == 0) { | |
5f39d397 | 2106 | struct btrfs_disk_key disk_key; |
e66f709b | 2107 | orig_slot += left_nr; |
5f39d397 | 2108 | btrfs_node_key(mid, &disk_key, 0); |
0e82bcfe DS |
2109 | ret = tree_mod_log_insert_key(parent, pslot, |
2110 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2111 | BUG_ON(ret < 0); | |
5f39d397 CM |
2112 | btrfs_set_node_key(parent, &disk_key, pslot); |
2113 | btrfs_mark_buffer_dirty(parent); | |
2114 | if (btrfs_header_nritems(left) > orig_slot) { | |
2115 | path->nodes[level] = left; | |
e66f709b CM |
2116 | path->slots[level + 1] -= 1; |
2117 | path->slots[level] = orig_slot; | |
925baedd | 2118 | btrfs_tree_unlock(mid); |
5f39d397 | 2119 | free_extent_buffer(mid); |
e66f709b CM |
2120 | } else { |
2121 | orig_slot -= | |
5f39d397 | 2122 | btrfs_header_nritems(left); |
e66f709b | 2123 | path->slots[level] = orig_slot; |
925baedd | 2124 | btrfs_tree_unlock(left); |
5f39d397 | 2125 | free_extent_buffer(left); |
e66f709b | 2126 | } |
e66f709b CM |
2127 | return 0; |
2128 | } | |
925baedd | 2129 | btrfs_tree_unlock(left); |
5f39d397 | 2130 | free_extent_buffer(left); |
e66f709b | 2131 | } |
4b231ae4 | 2132 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
2133 | if (IS_ERR(right)) |
2134 | right = NULL; | |
e66f709b CM |
2135 | |
2136 | /* | |
2137 | * then try to empty the right most buffer into the middle | |
2138 | */ | |
5f39d397 | 2139 | if (right) { |
33ade1f8 | 2140 | u32 right_nr; |
b4ce94de | 2141 | |
925baedd | 2142 | btrfs_tree_lock(right); |
8bead258 | 2143 | btrfs_set_lock_blocking_write(right); |
b4ce94de | 2144 | |
5f39d397 | 2145 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2146 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2147 | wret = 1; |
2148 | } else { | |
5f39d397 CM |
2149 | ret = btrfs_cow_block(trans, root, right, |
2150 | parent, pslot + 1, | |
9fa8cfe7 | 2151 | &right); |
54aa1f4d CM |
2152 | if (ret) |
2153 | wret = 1; | |
2154 | else { | |
55d32ed8 | 2155 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 2156 | } |
33ade1f8 | 2157 | } |
e66f709b CM |
2158 | if (wret < 0) |
2159 | ret = wret; | |
2160 | if (wret == 0) { | |
5f39d397 CM |
2161 | struct btrfs_disk_key disk_key; |
2162 | ||
2163 | btrfs_node_key(right, &disk_key, 0); | |
0e82bcfe DS |
2164 | ret = tree_mod_log_insert_key(parent, pslot + 1, |
2165 | MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
2166 | BUG_ON(ret < 0); | |
5f39d397 CM |
2167 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2168 | btrfs_mark_buffer_dirty(parent); | |
2169 | ||
2170 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2171 | path->nodes[level] = right; | |
e66f709b CM |
2172 | path->slots[level + 1] += 1; |
2173 | path->slots[level] = orig_slot - | |
5f39d397 | 2174 | btrfs_header_nritems(mid); |
925baedd | 2175 | btrfs_tree_unlock(mid); |
5f39d397 | 2176 | free_extent_buffer(mid); |
e66f709b | 2177 | } else { |
925baedd | 2178 | btrfs_tree_unlock(right); |
5f39d397 | 2179 | free_extent_buffer(right); |
e66f709b | 2180 | } |
e66f709b CM |
2181 | return 0; |
2182 | } | |
925baedd | 2183 | btrfs_tree_unlock(right); |
5f39d397 | 2184 | free_extent_buffer(right); |
e66f709b | 2185 | } |
e66f709b CM |
2186 | return 1; |
2187 | } | |
2188 | ||
3c69faec | 2189 | /* |
d352ac68 CM |
2190 | * readahead one full node of leaves, finding things that are close |
2191 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2192 | */ |
2ff7e61e | 2193 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2194 | struct btrfs_path *path, |
2195 | int level, int slot, u64 objectid) | |
3c69faec | 2196 | { |
5f39d397 | 2197 | struct extent_buffer *node; |
01f46658 | 2198 | struct btrfs_disk_key disk_key; |
3c69faec | 2199 | u32 nritems; |
3c69faec | 2200 | u64 search; |
a7175319 | 2201 | u64 target; |
6b80053d | 2202 | u64 nread = 0; |
5f39d397 | 2203 | struct extent_buffer *eb; |
6b80053d CM |
2204 | u32 nr; |
2205 | u32 blocksize; | |
2206 | u32 nscan = 0; | |
db94535d | 2207 | |
a6b6e75e | 2208 | if (level != 1) |
6702ed49 CM |
2209 | return; |
2210 | ||
2211 | if (!path->nodes[level]) | |
3c69faec CM |
2212 | return; |
2213 | ||
5f39d397 | 2214 | node = path->nodes[level]; |
925baedd | 2215 | |
3c69faec | 2216 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2217 | blocksize = fs_info->nodesize; |
2218 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2219 | if (eb) { |
2220 | free_extent_buffer(eb); | |
3c69faec CM |
2221 | return; |
2222 | } | |
2223 | ||
a7175319 | 2224 | target = search; |
6b80053d | 2225 | |
5f39d397 | 2226 | nritems = btrfs_header_nritems(node); |
6b80053d | 2227 | nr = slot; |
25b8b936 | 2228 | |
d397712b | 2229 | while (1) { |
e4058b54 | 2230 | if (path->reada == READA_BACK) { |
6b80053d CM |
2231 | if (nr == 0) |
2232 | break; | |
2233 | nr--; | |
e4058b54 | 2234 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2235 | nr++; |
2236 | if (nr >= nritems) | |
2237 | break; | |
3c69faec | 2238 | } |
e4058b54 | 2239 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2240 | btrfs_node_key(node, &disk_key, nr); |
2241 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2242 | break; | |
2243 | } | |
6b80053d | 2244 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2245 | if ((search <= target && target - search <= 65536) || |
2246 | (search > target && search - target <= 65536)) { | |
2ff7e61e | 2247 | readahead_tree_block(fs_info, search); |
6b80053d CM |
2248 | nread += blocksize; |
2249 | } | |
2250 | nscan++; | |
a7175319 | 2251 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2252 | break; |
3c69faec CM |
2253 | } |
2254 | } | |
925baedd | 2255 | |
2ff7e61e | 2256 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2257 | struct btrfs_path *path, int level) |
b4ce94de CM |
2258 | { |
2259 | int slot; | |
2260 | int nritems; | |
2261 | struct extent_buffer *parent; | |
2262 | struct extent_buffer *eb; | |
2263 | u64 gen; | |
2264 | u64 block1 = 0; | |
2265 | u64 block2 = 0; | |
b4ce94de | 2266 | |
8c594ea8 | 2267 | parent = path->nodes[level + 1]; |
b4ce94de | 2268 | if (!parent) |
0b08851f | 2269 | return; |
b4ce94de CM |
2270 | |
2271 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2272 | slot = path->slots[level + 1]; |
b4ce94de CM |
2273 | |
2274 | if (slot > 0) { | |
2275 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2276 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2277 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2278 | /* |
2279 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2280 | * don't want to return eagain here. That will loop | |
2281 | * forever | |
2282 | */ | |
2283 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2284 | block1 = 0; |
2285 | free_extent_buffer(eb); | |
2286 | } | |
8c594ea8 | 2287 | if (slot + 1 < nritems) { |
b4ce94de CM |
2288 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2289 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2290 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2291 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2292 | block2 = 0; |
2293 | free_extent_buffer(eb); | |
2294 | } | |
8c594ea8 | 2295 | |
0b08851f | 2296 | if (block1) |
2ff7e61e | 2297 | readahead_tree_block(fs_info, block1); |
0b08851f | 2298 | if (block2) |
2ff7e61e | 2299 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2300 | } |
2301 | ||
2302 | ||
d352ac68 | 2303 | /* |
d397712b CM |
2304 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2305 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2306 | * operations on the tree might require changing key pointers higher up in the | |
2307 | * tree. | |
d352ac68 | 2308 | * |
d397712b CM |
2309 | * callers might also have set path->keep_locks, which tells this code to keep |
2310 | * the lock if the path points to the last slot in the block. This is part of | |
2311 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2312 | * |
d397712b CM |
2313 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2314 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2315 | */ |
e02119d5 | 2316 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2317 | int lowest_unlock, int min_write_lock_level, |
2318 | int *write_lock_level) | |
925baedd CM |
2319 | { |
2320 | int i; | |
2321 | int skip_level = level; | |
051e1b9f | 2322 | int no_skips = 0; |
925baedd CM |
2323 | struct extent_buffer *t; |
2324 | ||
2325 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2326 | if (!path->nodes[i]) | |
2327 | break; | |
2328 | if (!path->locks[i]) | |
2329 | break; | |
051e1b9f | 2330 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2331 | skip_level = i + 1; |
2332 | continue; | |
2333 | } | |
051e1b9f | 2334 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2335 | u32 nritems; |
2336 | t = path->nodes[i]; | |
2337 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2338 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2339 | skip_level = i + 1; |
2340 | continue; | |
2341 | } | |
2342 | } | |
051e1b9f CM |
2343 | if (skip_level < i && i >= lowest_unlock) |
2344 | no_skips = 1; | |
2345 | ||
925baedd | 2346 | t = path->nodes[i]; |
d80bb3f9 | 2347 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 2348 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2349 | path->locks[i] = 0; |
f7c79f30 CM |
2350 | if (write_lock_level && |
2351 | i > min_write_lock_level && | |
2352 | i <= *write_lock_level) { | |
2353 | *write_lock_level = i - 1; | |
2354 | } | |
925baedd CM |
2355 | } |
2356 | } | |
2357 | } | |
2358 | ||
b4ce94de CM |
2359 | /* |
2360 | * This releases any locks held in the path starting at level and | |
2361 | * going all the way up to the root. | |
2362 | * | |
2363 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2364 | * corner cases, such as COW of the block at slot zero in the node. This | |
2365 | * ignores those rules, and it should only be called when there are no | |
2366 | * more updates to be done higher up in the tree. | |
2367 | */ | |
2368 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2369 | { | |
2370 | int i; | |
2371 | ||
09a2a8f9 | 2372 | if (path->keep_locks) |
b4ce94de CM |
2373 | return; |
2374 | ||
2375 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2376 | if (!path->nodes[i]) | |
12f4dacc | 2377 | continue; |
b4ce94de | 2378 | if (!path->locks[i]) |
12f4dacc | 2379 | continue; |
bd681513 | 2380 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2381 | path->locks[i] = 0; |
2382 | } | |
2383 | } | |
2384 | ||
c8c42864 CM |
2385 | /* |
2386 | * helper function for btrfs_search_slot. The goal is to find a block | |
2387 | * in cache without setting the path to blocking. If we find the block | |
2388 | * we return zero and the path is unchanged. | |
2389 | * | |
2390 | * If we can't find the block, we set the path blocking and do some | |
2391 | * reada. -EAGAIN is returned and the search must be repeated. | |
2392 | */ | |
2393 | static int | |
d07b8528 LB |
2394 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2395 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2396 | const struct btrfs_key *key) |
c8c42864 | 2397 | { |
0b246afa | 2398 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2399 | u64 blocknr; |
2400 | u64 gen; | |
c8c42864 CM |
2401 | struct extent_buffer *b = *eb_ret; |
2402 | struct extent_buffer *tmp; | |
581c1760 | 2403 | struct btrfs_key first_key; |
76a05b35 | 2404 | int ret; |
581c1760 | 2405 | int parent_level; |
c8c42864 CM |
2406 | |
2407 | blocknr = btrfs_node_blockptr(b, slot); | |
2408 | gen = btrfs_node_ptr_generation(b, slot); | |
581c1760 QW |
2409 | parent_level = btrfs_header_level(b); |
2410 | btrfs_node_key_to_cpu(b, &first_key, slot); | |
c8c42864 | 2411 | |
0b246afa | 2412 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2413 | if (tmp) { |
b9fab919 | 2414 | /* first we do an atomic uptodate check */ |
bdf7c00e | 2415 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
2416 | /* |
2417 | * Do extra check for first_key, eb can be stale due to | |
2418 | * being cached, read from scrub, or have multiple | |
2419 | * parents (shared tree blocks). | |
2420 | */ | |
e064d5e9 | 2421 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
2422 | parent_level - 1, &first_key, gen)) { |
2423 | free_extent_buffer(tmp); | |
2424 | return -EUCLEAN; | |
2425 | } | |
bdf7c00e JB |
2426 | *eb_ret = tmp; |
2427 | return 0; | |
2428 | } | |
2429 | ||
2430 | /* the pages were up to date, but we failed | |
2431 | * the generation number check. Do a full | |
2432 | * read for the generation number that is correct. | |
2433 | * We must do this without dropping locks so | |
2434 | * we can trust our generation number | |
2435 | */ | |
2436 | btrfs_set_path_blocking(p); | |
2437 | ||
2438 | /* now we're allowed to do a blocking uptodate check */ | |
581c1760 | 2439 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
2440 | if (!ret) { |
2441 | *eb_ret = tmp; | |
2442 | return 0; | |
cb44921a | 2443 | } |
bdf7c00e JB |
2444 | free_extent_buffer(tmp); |
2445 | btrfs_release_path(p); | |
2446 | return -EIO; | |
c8c42864 CM |
2447 | } |
2448 | ||
2449 | /* | |
2450 | * reduce lock contention at high levels | |
2451 | * of the btree by dropping locks before | |
76a05b35 CM |
2452 | * we read. Don't release the lock on the current |
2453 | * level because we need to walk this node to figure | |
2454 | * out which blocks to read. | |
c8c42864 | 2455 | */ |
8c594ea8 CM |
2456 | btrfs_unlock_up_safe(p, level + 1); |
2457 | btrfs_set_path_blocking(p); | |
2458 | ||
e4058b54 | 2459 | if (p->reada != READA_NONE) |
2ff7e61e | 2460 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2461 | |
76a05b35 | 2462 | ret = -EAGAIN; |
02a3307a | 2463 | tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1, |
581c1760 | 2464 | &first_key); |
64c043de | 2465 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2466 | /* |
2467 | * If the read above didn't mark this buffer up to date, | |
2468 | * it will never end up being up to date. Set ret to EIO now | |
2469 | * and give up so that our caller doesn't loop forever | |
2470 | * on our EAGAINs. | |
2471 | */ | |
e6a1d6fd | 2472 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 2473 | ret = -EIO; |
c8c42864 | 2474 | free_extent_buffer(tmp); |
c871b0f2 LB |
2475 | } else { |
2476 | ret = PTR_ERR(tmp); | |
76a05b35 | 2477 | } |
02a3307a LB |
2478 | |
2479 | btrfs_release_path(p); | |
76a05b35 | 2480 | return ret; |
c8c42864 CM |
2481 | } |
2482 | ||
2483 | /* | |
2484 | * helper function for btrfs_search_slot. This does all of the checks | |
2485 | * for node-level blocks and does any balancing required based on | |
2486 | * the ins_len. | |
2487 | * | |
2488 | * If no extra work was required, zero is returned. If we had to | |
2489 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2490 | * start over | |
2491 | */ | |
2492 | static int | |
2493 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2494 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2495 | struct extent_buffer *b, int level, int ins_len, |
2496 | int *write_lock_level) | |
c8c42864 | 2497 | { |
0b246afa | 2498 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2499 | int ret; |
0b246afa | 2500 | |
c8c42864 | 2501 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2502 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2503 | int sret; |
2504 | ||
bd681513 CM |
2505 | if (*write_lock_level < level + 1) { |
2506 | *write_lock_level = level + 1; | |
2507 | btrfs_release_path(p); | |
2508 | goto again; | |
2509 | } | |
2510 | ||
c8c42864 | 2511 | btrfs_set_path_blocking(p); |
2ff7e61e | 2512 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2513 | sret = split_node(trans, root, p, level); |
c8c42864 CM |
2514 | |
2515 | BUG_ON(sret > 0); | |
2516 | if (sret) { | |
2517 | ret = sret; | |
2518 | goto done; | |
2519 | } | |
2520 | b = p->nodes[level]; | |
2521 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2522 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2523 | int sret; |
2524 | ||
bd681513 CM |
2525 | if (*write_lock_level < level + 1) { |
2526 | *write_lock_level = level + 1; | |
2527 | btrfs_release_path(p); | |
2528 | goto again; | |
2529 | } | |
2530 | ||
c8c42864 | 2531 | btrfs_set_path_blocking(p); |
2ff7e61e | 2532 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2533 | sret = balance_level(trans, root, p, level); |
c8c42864 CM |
2534 | |
2535 | if (sret) { | |
2536 | ret = sret; | |
2537 | goto done; | |
2538 | } | |
2539 | b = p->nodes[level]; | |
2540 | if (!b) { | |
b3b4aa74 | 2541 | btrfs_release_path(p); |
c8c42864 CM |
2542 | goto again; |
2543 | } | |
2544 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2545 | } | |
2546 | return 0; | |
2547 | ||
2548 | again: | |
2549 | ret = -EAGAIN; | |
2550 | done: | |
2551 | return ret; | |
2552 | } | |
2553 | ||
310712b2 | 2554 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2555 | int level, int *prev_cmp, int *slot) |
2556 | { | |
2557 | if (*prev_cmp != 0) { | |
a74b35ec | 2558 | *prev_cmp = btrfs_bin_search(b, key, level, slot); |
d7396f07 FDBM |
2559 | return *prev_cmp; |
2560 | } | |
2561 | ||
d7396f07 FDBM |
2562 | *slot = 0; |
2563 | ||
2564 | return 0; | |
2565 | } | |
2566 | ||
381cf658 | 2567 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2568 | u64 iobjectid, u64 ioff, u8 key_type, |
2569 | struct btrfs_key *found_key) | |
2570 | { | |
2571 | int ret; | |
2572 | struct btrfs_key key; | |
2573 | struct extent_buffer *eb; | |
381cf658 DS |
2574 | |
2575 | ASSERT(path); | |
1d4c08e0 | 2576 | ASSERT(found_key); |
e33d5c3d KN |
2577 | |
2578 | key.type = key_type; | |
2579 | key.objectid = iobjectid; | |
2580 | key.offset = ioff; | |
2581 | ||
2582 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2583 | if (ret < 0) |
e33d5c3d KN |
2584 | return ret; |
2585 | ||
2586 | eb = path->nodes[0]; | |
2587 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2588 | ret = btrfs_next_leaf(fs_root, path); | |
2589 | if (ret) | |
2590 | return ret; | |
2591 | eb = path->nodes[0]; | |
2592 | } | |
2593 | ||
2594 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2595 | if (found_key->type != key.type || | |
2596 | found_key->objectid != key.objectid) | |
2597 | return 1; | |
2598 | ||
2599 | return 0; | |
2600 | } | |
2601 | ||
1fc28d8e LB |
2602 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
2603 | struct btrfs_path *p, | |
2604 | int write_lock_level) | |
2605 | { | |
2606 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2607 | struct extent_buffer *b; | |
2608 | int root_lock; | |
2609 | int level = 0; | |
2610 | ||
2611 | /* We try very hard to do read locks on the root */ | |
2612 | root_lock = BTRFS_READ_LOCK; | |
2613 | ||
2614 | if (p->search_commit_root) { | |
be6821f8 FM |
2615 | /* |
2616 | * The commit roots are read only so we always do read locks, | |
2617 | * and we always must hold the commit_root_sem when doing | |
2618 | * searches on them, the only exception is send where we don't | |
2619 | * want to block transaction commits for a long time, so | |
2620 | * we need to clone the commit root in order to avoid races | |
2621 | * with transaction commits that create a snapshot of one of | |
2622 | * the roots used by a send operation. | |
2623 | */ | |
2624 | if (p->need_commit_sem) { | |
1fc28d8e | 2625 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 2626 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 2627 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
2628 | if (!b) |
2629 | return ERR_PTR(-ENOMEM); | |
2630 | ||
2631 | } else { | |
2632 | b = root->commit_root; | |
2633 | extent_buffer_get(b); | |
2634 | } | |
2635 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
2636 | /* |
2637 | * Ensure that all callers have set skip_locking when | |
2638 | * p->search_commit_root = 1. | |
2639 | */ | |
2640 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
2641 | |
2642 | goto out; | |
2643 | } | |
2644 | ||
2645 | if (p->skip_locking) { | |
2646 | b = btrfs_root_node(root); | |
2647 | level = btrfs_header_level(b); | |
2648 | goto out; | |
2649 | } | |
2650 | ||
2651 | /* | |
662c653b LB |
2652 | * If the level is set to maximum, we can skip trying to get the read |
2653 | * lock. | |
1fc28d8e | 2654 | */ |
662c653b LB |
2655 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
2656 | /* | |
2657 | * We don't know the level of the root node until we actually | |
2658 | * have it read locked | |
2659 | */ | |
2660 | b = btrfs_read_lock_root_node(root); | |
2661 | level = btrfs_header_level(b); | |
2662 | if (level > write_lock_level) | |
2663 | goto out; | |
2664 | ||
2665 | /* Whoops, must trade for write lock */ | |
2666 | btrfs_tree_read_unlock(b); | |
2667 | free_extent_buffer(b); | |
2668 | } | |
1fc28d8e | 2669 | |
1fc28d8e LB |
2670 | b = btrfs_lock_root_node(root); |
2671 | root_lock = BTRFS_WRITE_LOCK; | |
2672 | ||
2673 | /* The level might have changed, check again */ | |
2674 | level = btrfs_header_level(b); | |
2675 | ||
2676 | out: | |
2677 | p->nodes[level] = b; | |
2678 | if (!p->skip_locking) | |
2679 | p->locks[level] = root_lock; | |
2680 | /* | |
2681 | * Callers are responsible for dropping b's references. | |
2682 | */ | |
2683 | return b; | |
2684 | } | |
2685 | ||
2686 | ||
74123bd7 | 2687 | /* |
4271ecea NB |
2688 | * btrfs_search_slot - look for a key in a tree and perform necessary |
2689 | * modifications to preserve tree invariants. | |
74123bd7 | 2690 | * |
4271ecea NB |
2691 | * @trans: Handle of transaction, used when modifying the tree |
2692 | * @p: Holds all btree nodes along the search path | |
2693 | * @root: The root node of the tree | |
2694 | * @key: The key we are looking for | |
2695 | * @ins_len: Indicates purpose of search, for inserts it is 1, for | |
2696 | * deletions it's -1. 0 for plain searches | |
2697 | * @cow: boolean should CoW operations be performed. Must always be 1 | |
2698 | * when modifying the tree. | |
97571fd0 | 2699 | * |
4271ecea NB |
2700 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
2701 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
2702 | * | |
2703 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
2704 | * of the path (level 0) | |
2705 | * | |
2706 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
2707 | * points to the slot where it should be inserted | |
2708 | * | |
2709 | * If an error is encountered while searching the tree a negative error number | |
2710 | * is returned | |
74123bd7 | 2711 | */ |
310712b2 OS |
2712 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2713 | const struct btrfs_key *key, struct btrfs_path *p, | |
2714 | int ins_len, int cow) | |
be0e5c09 | 2715 | { |
5f39d397 | 2716 | struct extent_buffer *b; |
be0e5c09 CM |
2717 | int slot; |
2718 | int ret; | |
33c66f43 | 2719 | int err; |
be0e5c09 | 2720 | int level; |
925baedd | 2721 | int lowest_unlock = 1; |
bd681513 CM |
2722 | /* everything at write_lock_level or lower must be write locked */ |
2723 | int write_lock_level = 0; | |
9f3a7427 | 2724 | u8 lowest_level = 0; |
f7c79f30 | 2725 | int min_write_lock_level; |
d7396f07 | 2726 | int prev_cmp; |
9f3a7427 | 2727 | |
6702ed49 | 2728 | lowest_level = p->lowest_level; |
323ac95b | 2729 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2730 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2731 | BUG_ON(!cow && ins_len); |
25179201 | 2732 | |
bd681513 | 2733 | if (ins_len < 0) { |
925baedd | 2734 | lowest_unlock = 2; |
65b51a00 | 2735 | |
bd681513 CM |
2736 | /* when we are removing items, we might have to go up to level |
2737 | * two as we update tree pointers Make sure we keep write | |
2738 | * for those levels as well | |
2739 | */ | |
2740 | write_lock_level = 2; | |
2741 | } else if (ins_len > 0) { | |
2742 | /* | |
2743 | * for inserting items, make sure we have a write lock on | |
2744 | * level 1 so we can update keys | |
2745 | */ | |
2746 | write_lock_level = 1; | |
2747 | } | |
2748 | ||
2749 | if (!cow) | |
2750 | write_lock_level = -1; | |
2751 | ||
09a2a8f9 | 2752 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2753 | write_lock_level = BTRFS_MAX_LEVEL; |
2754 | ||
f7c79f30 CM |
2755 | min_write_lock_level = write_lock_level; |
2756 | ||
bb803951 | 2757 | again: |
d7396f07 | 2758 | prev_cmp = -1; |
1fc28d8e | 2759 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2760 | if (IS_ERR(b)) { |
2761 | ret = PTR_ERR(b); | |
2762 | goto done; | |
2763 | } | |
925baedd | 2764 | |
eb60ceac | 2765 | while (b) { |
5f39d397 | 2766 | level = btrfs_header_level(b); |
65b51a00 CM |
2767 | |
2768 | /* | |
2769 | * setup the path here so we can release it under lock | |
2770 | * contention with the cow code | |
2771 | */ | |
02217ed2 | 2772 | if (cow) { |
9ea2c7c9 NB |
2773 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2774 | ||
c8c42864 CM |
2775 | /* |
2776 | * if we don't really need to cow this block | |
2777 | * then we don't want to set the path blocking, | |
2778 | * so we test it here | |
2779 | */ | |
64c12921 JM |
2780 | if (!should_cow_block(trans, root, b)) { |
2781 | trans->dirty = true; | |
65b51a00 | 2782 | goto cow_done; |
64c12921 | 2783 | } |
5d4f98a2 | 2784 | |
bd681513 CM |
2785 | /* |
2786 | * must have write locks on this node and the | |
2787 | * parent | |
2788 | */ | |
5124e00e JB |
2789 | if (level > write_lock_level || |
2790 | (level + 1 > write_lock_level && | |
2791 | level + 1 < BTRFS_MAX_LEVEL && | |
2792 | p->nodes[level + 1])) { | |
bd681513 CM |
2793 | write_lock_level = level + 1; |
2794 | btrfs_release_path(p); | |
2795 | goto again; | |
2796 | } | |
2797 | ||
160f4089 | 2798 | btrfs_set_path_blocking(p); |
9ea2c7c9 NB |
2799 | if (last_level) |
2800 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
2801 | &b); | |
2802 | else | |
2803 | err = btrfs_cow_block(trans, root, b, | |
2804 | p->nodes[level + 1], | |
2805 | p->slots[level + 1], &b); | |
33c66f43 | 2806 | if (err) { |
33c66f43 | 2807 | ret = err; |
65b51a00 | 2808 | goto done; |
54aa1f4d | 2809 | } |
02217ed2 | 2810 | } |
65b51a00 | 2811 | cow_done: |
eb60ceac | 2812 | p->nodes[level] = b; |
52398340 LB |
2813 | /* |
2814 | * Leave path with blocking locks to avoid massive | |
2815 | * lock context switch, this is made on purpose. | |
2816 | */ | |
b4ce94de CM |
2817 | |
2818 | /* | |
2819 | * we have a lock on b and as long as we aren't changing | |
2820 | * the tree, there is no way to for the items in b to change. | |
2821 | * It is safe to drop the lock on our parent before we | |
2822 | * go through the expensive btree search on b. | |
2823 | * | |
eb653de1 FDBM |
2824 | * If we're inserting or deleting (ins_len != 0), then we might |
2825 | * be changing slot zero, which may require changing the parent. | |
2826 | * So, we can't drop the lock until after we know which slot | |
2827 | * we're operating on. | |
b4ce94de | 2828 | */ |
eb653de1 FDBM |
2829 | if (!ins_len && !p->keep_locks) { |
2830 | int u = level + 1; | |
2831 | ||
2832 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2833 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2834 | p->locks[u] = 0; | |
2835 | } | |
2836 | } | |
b4ce94de | 2837 | |
d7396f07 | 2838 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2839 | if (ret < 0) |
2840 | goto done; | |
b4ce94de | 2841 | |
5f39d397 | 2842 | if (level != 0) { |
33c66f43 YZ |
2843 | int dec = 0; |
2844 | if (ret && slot > 0) { | |
2845 | dec = 1; | |
be0e5c09 | 2846 | slot -= 1; |
33c66f43 | 2847 | } |
be0e5c09 | 2848 | p->slots[level] = slot; |
33c66f43 | 2849 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2850 | ins_len, &write_lock_level); |
33c66f43 | 2851 | if (err == -EAGAIN) |
c8c42864 | 2852 | goto again; |
33c66f43 YZ |
2853 | if (err) { |
2854 | ret = err; | |
c8c42864 | 2855 | goto done; |
33c66f43 | 2856 | } |
c8c42864 CM |
2857 | b = p->nodes[level]; |
2858 | slot = p->slots[level]; | |
b4ce94de | 2859 | |
bd681513 CM |
2860 | /* |
2861 | * slot 0 is special, if we change the key | |
2862 | * we have to update the parent pointer | |
2863 | * which means we must have a write lock | |
2864 | * on the parent | |
2865 | */ | |
eb653de1 | 2866 | if (slot == 0 && ins_len && |
bd681513 CM |
2867 | write_lock_level < level + 1) { |
2868 | write_lock_level = level + 1; | |
2869 | btrfs_release_path(p); | |
2870 | goto again; | |
2871 | } | |
2872 | ||
f7c79f30 CM |
2873 | unlock_up(p, level, lowest_unlock, |
2874 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2875 | |
925baedd | 2876 | if (level == lowest_level) { |
33c66f43 YZ |
2877 | if (dec) |
2878 | p->slots[level]++; | |
5b21f2ed | 2879 | goto done; |
925baedd | 2880 | } |
ca7a79ad | 2881 | |
d07b8528 | 2882 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2883 | slot, key); |
33c66f43 | 2884 | if (err == -EAGAIN) |
c8c42864 | 2885 | goto again; |
33c66f43 YZ |
2886 | if (err) { |
2887 | ret = err; | |
76a05b35 | 2888 | goto done; |
33c66f43 | 2889 | } |
76a05b35 | 2890 | |
b4ce94de | 2891 | if (!p->skip_locking) { |
bd681513 CM |
2892 | level = btrfs_header_level(b); |
2893 | if (level <= write_lock_level) { | |
2894 | err = btrfs_try_tree_write_lock(b); | |
2895 | if (!err) { | |
2896 | btrfs_set_path_blocking(p); | |
2897 | btrfs_tree_lock(b); | |
bd681513 CM |
2898 | } |
2899 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2900 | } else { | |
f82c458a | 2901 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2902 | if (!err) { |
2903 | btrfs_set_path_blocking(p); | |
2904 | btrfs_tree_read_lock(b); | |
bd681513 CM |
2905 | } |
2906 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2907 | } |
bd681513 | 2908 | p->nodes[level] = b; |
b4ce94de | 2909 | } |
be0e5c09 CM |
2910 | } else { |
2911 | p->slots[level] = slot; | |
87b29b20 | 2912 | if (ins_len > 0 && |
e902baac | 2913 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
2914 | if (write_lock_level < 1) { |
2915 | write_lock_level = 1; | |
2916 | btrfs_release_path(p); | |
2917 | goto again; | |
2918 | } | |
2919 | ||
b4ce94de | 2920 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2921 | err = split_leaf(trans, root, key, |
2922 | p, ins_len, ret == 0); | |
b4ce94de | 2923 | |
33c66f43 YZ |
2924 | BUG_ON(err > 0); |
2925 | if (err) { | |
2926 | ret = err; | |
65b51a00 CM |
2927 | goto done; |
2928 | } | |
5c680ed6 | 2929 | } |
459931ec | 2930 | if (!p->search_for_split) |
f7c79f30 | 2931 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2932 | min_write_lock_level, NULL); |
65b51a00 | 2933 | goto done; |
be0e5c09 CM |
2934 | } |
2935 | } | |
65b51a00 CM |
2936 | ret = 1; |
2937 | done: | |
b4ce94de CM |
2938 | /* |
2939 | * we don't really know what they plan on doing with the path | |
2940 | * from here on, so for now just mark it as blocking | |
2941 | */ | |
b9473439 CM |
2942 | if (!p->leave_spinning) |
2943 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2944 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2945 | btrfs_release_path(p); |
65b51a00 | 2946 | return ret; |
be0e5c09 CM |
2947 | } |
2948 | ||
5d9e75c4 JS |
2949 | /* |
2950 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2951 | * current state of the tree together with the operations recorded in the tree | |
2952 | * modification log to search for the key in a previous version of this tree, as | |
2953 | * denoted by the time_seq parameter. | |
2954 | * | |
2955 | * Naturally, there is no support for insert, delete or cow operations. | |
2956 | * | |
2957 | * The resulting path and return value will be set up as if we called | |
2958 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2959 | */ | |
310712b2 | 2960 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2961 | struct btrfs_path *p, u64 time_seq) |
2962 | { | |
0b246afa | 2963 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2964 | struct extent_buffer *b; |
2965 | int slot; | |
2966 | int ret; | |
2967 | int err; | |
2968 | int level; | |
2969 | int lowest_unlock = 1; | |
2970 | u8 lowest_level = 0; | |
d4b4087c | 2971 | int prev_cmp = -1; |
5d9e75c4 JS |
2972 | |
2973 | lowest_level = p->lowest_level; | |
2974 | WARN_ON(p->nodes[0] != NULL); | |
2975 | ||
2976 | if (p->search_commit_root) { | |
2977 | BUG_ON(time_seq); | |
2978 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2979 | } | |
2980 | ||
2981 | again: | |
5d9e75c4 | 2982 | b = get_old_root(root, time_seq); |
315bed43 NB |
2983 | if (!b) { |
2984 | ret = -EIO; | |
2985 | goto done; | |
2986 | } | |
5d9e75c4 | 2987 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2988 | p->locks[level] = BTRFS_READ_LOCK; |
2989 | ||
2990 | while (b) { | |
2991 | level = btrfs_header_level(b); | |
2992 | p->nodes[level] = b; | |
5d9e75c4 JS |
2993 | |
2994 | /* | |
2995 | * we have a lock on b and as long as we aren't changing | |
2996 | * the tree, there is no way to for the items in b to change. | |
2997 | * It is safe to drop the lock on our parent before we | |
2998 | * go through the expensive btree search on b. | |
2999 | */ | |
3000 | btrfs_unlock_up_safe(p, level + 1); | |
3001 | ||
d4b4087c | 3002 | /* |
01327610 | 3003 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
3004 | * time. |
3005 | */ | |
3006 | prev_cmp = -1; | |
d7396f07 | 3007 | ret = key_search(b, key, level, &prev_cmp, &slot); |
cbca7d59 FM |
3008 | if (ret < 0) |
3009 | goto done; | |
5d9e75c4 JS |
3010 | |
3011 | if (level != 0) { | |
3012 | int dec = 0; | |
3013 | if (ret && slot > 0) { | |
3014 | dec = 1; | |
3015 | slot -= 1; | |
3016 | } | |
3017 | p->slots[level] = slot; | |
3018 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3019 | ||
3020 | if (level == lowest_level) { | |
3021 | if (dec) | |
3022 | p->slots[level]++; | |
3023 | goto done; | |
3024 | } | |
3025 | ||
d07b8528 | 3026 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3027 | slot, key); |
5d9e75c4 JS |
3028 | if (err == -EAGAIN) |
3029 | goto again; | |
3030 | if (err) { | |
3031 | ret = err; | |
3032 | goto done; | |
3033 | } | |
3034 | ||
3035 | level = btrfs_header_level(b); | |
f82c458a | 3036 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3037 | if (!err) { |
3038 | btrfs_set_path_blocking(p); | |
3039 | btrfs_tree_read_lock(b); | |
5d9e75c4 | 3040 | } |
0b246afa | 3041 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3042 | if (!b) { |
3043 | ret = -ENOMEM; | |
3044 | goto done; | |
3045 | } | |
5d9e75c4 JS |
3046 | p->locks[level] = BTRFS_READ_LOCK; |
3047 | p->nodes[level] = b; | |
5d9e75c4 JS |
3048 | } else { |
3049 | p->slots[level] = slot; | |
3050 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3051 | goto done; | |
3052 | } | |
3053 | } | |
3054 | ret = 1; | |
3055 | done: | |
3056 | if (!p->leave_spinning) | |
3057 | btrfs_set_path_blocking(p); | |
3058 | if (ret < 0) | |
3059 | btrfs_release_path(p); | |
3060 | ||
3061 | return ret; | |
3062 | } | |
3063 | ||
2f38b3e1 AJ |
3064 | /* |
3065 | * helper to use instead of search slot if no exact match is needed but | |
3066 | * instead the next or previous item should be returned. | |
3067 | * When find_higher is true, the next higher item is returned, the next lower | |
3068 | * otherwise. | |
3069 | * When return_any and find_higher are both true, and no higher item is found, | |
3070 | * return the next lower instead. | |
3071 | * When return_any is true and find_higher is false, and no lower item is found, | |
3072 | * return the next higher instead. | |
3073 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3074 | * < 0 on error | |
3075 | */ | |
3076 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3077 | const struct btrfs_key *key, |
3078 | struct btrfs_path *p, int find_higher, | |
3079 | int return_any) | |
2f38b3e1 AJ |
3080 | { |
3081 | int ret; | |
3082 | struct extent_buffer *leaf; | |
3083 | ||
3084 | again: | |
3085 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3086 | if (ret <= 0) | |
3087 | return ret; | |
3088 | /* | |
3089 | * a return value of 1 means the path is at the position where the | |
3090 | * item should be inserted. Normally this is the next bigger item, | |
3091 | * but in case the previous item is the last in a leaf, path points | |
3092 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3093 | * item. | |
3094 | */ | |
3095 | leaf = p->nodes[0]; | |
3096 | ||
3097 | if (find_higher) { | |
3098 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3099 | ret = btrfs_next_leaf(root, p); | |
3100 | if (ret <= 0) | |
3101 | return ret; | |
3102 | if (!return_any) | |
3103 | return 1; | |
3104 | /* | |
3105 | * no higher item found, return the next | |
3106 | * lower instead | |
3107 | */ | |
3108 | return_any = 0; | |
3109 | find_higher = 0; | |
3110 | btrfs_release_path(p); | |
3111 | goto again; | |
3112 | } | |
3113 | } else { | |
e6793769 AJ |
3114 | if (p->slots[0] == 0) { |
3115 | ret = btrfs_prev_leaf(root, p); | |
3116 | if (ret < 0) | |
3117 | return ret; | |
3118 | if (!ret) { | |
23c6bf6a FDBM |
3119 | leaf = p->nodes[0]; |
3120 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3121 | p->slots[0]--; | |
e6793769 | 3122 | return 0; |
2f38b3e1 | 3123 | } |
e6793769 AJ |
3124 | if (!return_any) |
3125 | return 1; | |
3126 | /* | |
3127 | * no lower item found, return the next | |
3128 | * higher instead | |
3129 | */ | |
3130 | return_any = 0; | |
3131 | find_higher = 1; | |
3132 | btrfs_release_path(p); | |
3133 | goto again; | |
3134 | } else { | |
2f38b3e1 AJ |
3135 | --p->slots[0]; |
3136 | } | |
3137 | } | |
3138 | return 0; | |
3139 | } | |
3140 | ||
74123bd7 CM |
3141 | /* |
3142 | * adjust the pointers going up the tree, starting at level | |
3143 | * making sure the right key of each node is points to 'key'. | |
3144 | * This is used after shifting pointers to the left, so it stops | |
3145 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3146 | * higher levels | |
aa5d6bed | 3147 | * |
74123bd7 | 3148 | */ |
b167fa91 | 3149 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 3150 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3151 | { |
3152 | int i; | |
5f39d397 | 3153 | struct extent_buffer *t; |
0e82bcfe | 3154 | int ret; |
5f39d397 | 3155 | |
234b63a0 | 3156 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3157 | int tslot = path->slots[i]; |
0e82bcfe | 3158 | |
eb60ceac | 3159 | if (!path->nodes[i]) |
be0e5c09 | 3160 | break; |
5f39d397 | 3161 | t = path->nodes[i]; |
0e82bcfe DS |
3162 | ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE, |
3163 | GFP_ATOMIC); | |
3164 | BUG_ON(ret < 0); | |
5f39d397 | 3165 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3166 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3167 | if (tslot != 0) |
3168 | break; | |
3169 | } | |
3170 | } | |
3171 | ||
31840ae1 ZY |
3172 | /* |
3173 | * update item key. | |
3174 | * | |
3175 | * This function isn't completely safe. It's the caller's responsibility | |
3176 | * that the new key won't break the order | |
3177 | */ | |
b7a0365e DD |
3178 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3179 | struct btrfs_path *path, | |
310712b2 | 3180 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3181 | { |
3182 | struct btrfs_disk_key disk_key; | |
3183 | struct extent_buffer *eb; | |
3184 | int slot; | |
3185 | ||
3186 | eb = path->nodes[0]; | |
3187 | slot = path->slots[0]; | |
3188 | if (slot > 0) { | |
3189 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
3190 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
3191 | btrfs_crit(fs_info, | |
3192 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3193 | slot, btrfs_disk_key_objectid(&disk_key), | |
3194 | btrfs_disk_key_type(&disk_key), | |
3195 | btrfs_disk_key_offset(&disk_key), | |
3196 | new_key->objectid, new_key->type, | |
3197 | new_key->offset); | |
3198 | btrfs_print_leaf(eb); | |
3199 | BUG(); | |
3200 | } | |
31840ae1 ZY |
3201 | } |
3202 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3203 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
3204 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
3205 | btrfs_crit(fs_info, | |
3206 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
3207 | slot, btrfs_disk_key_objectid(&disk_key), | |
3208 | btrfs_disk_key_type(&disk_key), | |
3209 | btrfs_disk_key_offset(&disk_key), | |
3210 | new_key->objectid, new_key->type, | |
3211 | new_key->offset); | |
3212 | btrfs_print_leaf(eb); | |
3213 | BUG(); | |
3214 | } | |
31840ae1 ZY |
3215 | } |
3216 | ||
3217 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3218 | btrfs_set_item_key(eb, &disk_key, slot); | |
3219 | btrfs_mark_buffer_dirty(eb); | |
3220 | if (slot == 0) | |
b167fa91 | 3221 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
3222 | } |
3223 | ||
74123bd7 CM |
3224 | /* |
3225 | * try to push data from one node into the next node left in the | |
79f95c82 | 3226 | * tree. |
aa5d6bed CM |
3227 | * |
3228 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3229 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3230 | */ |
98ed5174 | 3231 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 3232 | struct extent_buffer *dst, |
971a1f66 | 3233 | struct extent_buffer *src, int empty) |
be0e5c09 | 3234 | { |
d30a668f | 3235 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 3236 | int push_items = 0; |
bb803951 CM |
3237 | int src_nritems; |
3238 | int dst_nritems; | |
aa5d6bed | 3239 | int ret = 0; |
be0e5c09 | 3240 | |
5f39d397 CM |
3241 | src_nritems = btrfs_header_nritems(src); |
3242 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3243 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3244 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3245 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3246 | |
bce4eae9 | 3247 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3248 | return 1; |
3249 | ||
d397712b | 3250 | if (push_items <= 0) |
be0e5c09 CM |
3251 | return 1; |
3252 | ||
bce4eae9 | 3253 | if (empty) { |
971a1f66 | 3254 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3255 | if (push_items < src_nritems) { |
3256 | /* leave at least 8 pointers in the node if | |
3257 | * we aren't going to empty it | |
3258 | */ | |
3259 | if (src_nritems - push_items < 8) { | |
3260 | if (push_items <= 8) | |
3261 | return 1; | |
3262 | push_items -= 8; | |
3263 | } | |
3264 | } | |
3265 | } else | |
3266 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3267 | |
ed874f0d | 3268 | ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 3269 | if (ret) { |
66642832 | 3270 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3271 | return ret; |
3272 | } | |
5f39d397 CM |
3273 | copy_extent_buffer(dst, src, |
3274 | btrfs_node_key_ptr_offset(dst_nritems), | |
3275 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3276 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3277 | |
bb803951 | 3278 | if (push_items < src_nritems) { |
57911b8b | 3279 | /* |
bf1d3425 DS |
3280 | * Don't call tree_mod_log_insert_move here, key removal was |
3281 | * already fully logged by tree_mod_log_eb_copy above. | |
57911b8b | 3282 | */ |
5f39d397 CM |
3283 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3284 | btrfs_node_key_ptr_offset(push_items), | |
3285 | (src_nritems - push_items) * | |
3286 | sizeof(struct btrfs_key_ptr)); | |
3287 | } | |
3288 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3289 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3290 | btrfs_mark_buffer_dirty(src); | |
3291 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3292 | |
79f95c82 CM |
3293 | return ret; |
3294 | } | |
3295 | ||
3296 | /* | |
3297 | * try to push data from one node into the next node right in the | |
3298 | * tree. | |
3299 | * | |
3300 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3301 | * error, and > 0 if there was no room in the right hand block. | |
3302 | * | |
3303 | * this will only push up to 1/2 the contents of the left node over | |
3304 | */ | |
5f39d397 | 3305 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3306 | struct extent_buffer *dst, |
3307 | struct extent_buffer *src) | |
79f95c82 | 3308 | { |
55d32ed8 | 3309 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
3310 | int push_items = 0; |
3311 | int max_push; | |
3312 | int src_nritems; | |
3313 | int dst_nritems; | |
3314 | int ret = 0; | |
79f95c82 | 3315 | |
7bb86316 CM |
3316 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3317 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3318 | ||
5f39d397 CM |
3319 | src_nritems = btrfs_header_nritems(src); |
3320 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3321 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3322 | if (push_items <= 0) |
79f95c82 | 3323 | return 1; |
bce4eae9 | 3324 | |
d397712b | 3325 | if (src_nritems < 4) |
bce4eae9 | 3326 | return 1; |
79f95c82 CM |
3327 | |
3328 | max_push = src_nritems / 2 + 1; | |
3329 | /* don't try to empty the node */ | |
d397712b | 3330 | if (max_push >= src_nritems) |
79f95c82 | 3331 | return 1; |
252c38f0 | 3332 | |
79f95c82 CM |
3333 | if (max_push < push_items) |
3334 | push_items = max_push; | |
3335 | ||
bf1d3425 DS |
3336 | ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
3337 | BUG_ON(ret < 0); | |
5f39d397 CM |
3338 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3339 | btrfs_node_key_ptr_offset(0), | |
3340 | (dst_nritems) * | |
3341 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3342 | |
ed874f0d DS |
3343 | ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
3344 | push_items); | |
5de865ee | 3345 | if (ret) { |
66642832 | 3346 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3347 | return ret; |
3348 | } | |
5f39d397 CM |
3349 | copy_extent_buffer(dst, src, |
3350 | btrfs_node_key_ptr_offset(0), | |
3351 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3352 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3353 | |
5f39d397 CM |
3354 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3355 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3356 | |
5f39d397 CM |
3357 | btrfs_mark_buffer_dirty(src); |
3358 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3359 | |
aa5d6bed | 3360 | return ret; |
be0e5c09 CM |
3361 | } |
3362 | ||
97571fd0 CM |
3363 | /* |
3364 | * helper function to insert a new root level in the tree. | |
3365 | * A new node is allocated, and a single item is inserted to | |
3366 | * point to the existing root | |
aa5d6bed CM |
3367 | * |
3368 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3369 | */ |
d397712b | 3370 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3371 | struct btrfs_root *root, |
fdd99c72 | 3372 | struct btrfs_path *path, int level) |
5c680ed6 | 3373 | { |
0b246afa | 3374 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3375 | u64 lower_gen; |
5f39d397 CM |
3376 | struct extent_buffer *lower; |
3377 | struct extent_buffer *c; | |
925baedd | 3378 | struct extent_buffer *old; |
5f39d397 | 3379 | struct btrfs_disk_key lower_key; |
d9d19a01 | 3380 | int ret; |
5c680ed6 CM |
3381 | |
3382 | BUG_ON(path->nodes[level]); | |
3383 | BUG_ON(path->nodes[level-1] != root->node); | |
3384 | ||
7bb86316 CM |
3385 | lower = path->nodes[level-1]; |
3386 | if (level == 1) | |
3387 | btrfs_item_key(lower, &lower_key, 0); | |
3388 | else | |
3389 | btrfs_node_key(lower, &lower_key, 0); | |
3390 | ||
a6279470 FM |
3391 | c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level, |
3392 | root->node->start, 0); | |
5f39d397 CM |
3393 | if (IS_ERR(c)) |
3394 | return PTR_ERR(c); | |
925baedd | 3395 | |
0b246afa | 3396 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3397 | |
5f39d397 | 3398 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 3399 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3400 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3401 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3402 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3403 | |
3404 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3405 | |
5f39d397 | 3406 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3407 | |
925baedd | 3408 | old = root->node; |
d9d19a01 DS |
3409 | ret = tree_mod_log_insert_root(root->node, c, 0); |
3410 | BUG_ON(ret < 0); | |
240f62c8 | 3411 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3412 | |
3413 | /* the super has an extra ref to root->node */ | |
3414 | free_extent_buffer(old); | |
3415 | ||
0b86a832 | 3416 | add_root_to_dirty_list(root); |
5f39d397 CM |
3417 | extent_buffer_get(c); |
3418 | path->nodes[level] = c; | |
95449a16 | 3419 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3420 | path->slots[level] = 0; |
3421 | return 0; | |
3422 | } | |
3423 | ||
74123bd7 CM |
3424 | /* |
3425 | * worker function to insert a single pointer in a node. | |
3426 | * the node should have enough room for the pointer already | |
97571fd0 | 3427 | * |
74123bd7 CM |
3428 | * slot and level indicate where you want the key to go, and |
3429 | * blocknr is the block the key points to. | |
3430 | */ | |
143bede5 | 3431 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 3432 | struct btrfs_path *path, |
143bede5 | 3433 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3434 | int slot, int level) |
74123bd7 | 3435 | { |
5f39d397 | 3436 | struct extent_buffer *lower; |
74123bd7 | 3437 | int nritems; |
f3ea38da | 3438 | int ret; |
5c680ed6 CM |
3439 | |
3440 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3441 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3442 | lower = path->nodes[level]; |
3443 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3444 | BUG_ON(slot > nritems); |
6ad3cf6d | 3445 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 3446 | if (slot != nritems) { |
bf1d3425 DS |
3447 | if (level) { |
3448 | ret = tree_mod_log_insert_move(lower, slot + 1, slot, | |
a446a979 | 3449 | nritems - slot); |
bf1d3425 DS |
3450 | BUG_ON(ret < 0); |
3451 | } | |
5f39d397 CM |
3452 | memmove_extent_buffer(lower, |
3453 | btrfs_node_key_ptr_offset(slot + 1), | |
3454 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3455 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3456 | } |
c3e06965 | 3457 | if (level) { |
e09c2efe DS |
3458 | ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD, |
3459 | GFP_NOFS); | |
f3ea38da JS |
3460 | BUG_ON(ret < 0); |
3461 | } | |
5f39d397 | 3462 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3463 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3464 | WARN_ON(trans->transid == 0); |
3465 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3466 | btrfs_set_header_nritems(lower, nritems + 1); |
3467 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3468 | } |
3469 | ||
97571fd0 CM |
3470 | /* |
3471 | * split the node at the specified level in path in two. | |
3472 | * The path is corrected to point to the appropriate node after the split | |
3473 | * | |
3474 | * Before splitting this tries to make some room in the node by pushing | |
3475 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3476 | * |
3477 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3478 | */ |
e02119d5 CM |
3479 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3480 | struct btrfs_root *root, | |
3481 | struct btrfs_path *path, int level) | |
be0e5c09 | 3482 | { |
0b246afa | 3483 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3484 | struct extent_buffer *c; |
3485 | struct extent_buffer *split; | |
3486 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3487 | int mid; |
5c680ed6 | 3488 | int ret; |
7518a238 | 3489 | u32 c_nritems; |
eb60ceac | 3490 | |
5f39d397 | 3491 | c = path->nodes[level]; |
7bb86316 | 3492 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3493 | if (c == root->node) { |
d9abbf1c | 3494 | /* |
90f8d62e JS |
3495 | * trying to split the root, lets make a new one |
3496 | * | |
fdd99c72 | 3497 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3498 | * insert_new_root, because that root buffer will be kept as a |
3499 | * normal node. We are going to log removal of half of the | |
3500 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3501 | * tree lock on the buffer, which is why we cannot race with | |
3502 | * other tree_mod_log users. | |
d9abbf1c | 3503 | */ |
fdd99c72 | 3504 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3505 | if (ret) |
3506 | return ret; | |
b3612421 | 3507 | } else { |
e66f709b | 3508 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3509 | c = path->nodes[level]; |
3510 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3511 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3512 | return 0; |
54aa1f4d CM |
3513 | if (ret < 0) |
3514 | return ret; | |
be0e5c09 | 3515 | } |
e66f709b | 3516 | |
5f39d397 | 3517 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3518 | mid = (c_nritems + 1) / 2; |
3519 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3520 | |
a6279470 FM |
3521 | split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level, |
3522 | c->start, 0); | |
5f39d397 CM |
3523 | if (IS_ERR(split)) |
3524 | return PTR_ERR(split); | |
3525 | ||
0b246afa | 3526 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 3527 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3528 | |
ed874f0d | 3529 | ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3530 | if (ret) { |
66642832 | 3531 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3532 | return ret; |
3533 | } | |
5f39d397 CM |
3534 | copy_extent_buffer(split, c, |
3535 | btrfs_node_key_ptr_offset(0), | |
3536 | btrfs_node_key_ptr_offset(mid), | |
3537 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3538 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3539 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3540 | ret = 0; |
3541 | ||
5f39d397 CM |
3542 | btrfs_mark_buffer_dirty(c); |
3543 | btrfs_mark_buffer_dirty(split); | |
3544 | ||
6ad3cf6d | 3545 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 3546 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3547 | |
5de08d7d | 3548 | if (path->slots[level] >= mid) { |
5c680ed6 | 3549 | path->slots[level] -= mid; |
925baedd | 3550 | btrfs_tree_unlock(c); |
5f39d397 CM |
3551 | free_extent_buffer(c); |
3552 | path->nodes[level] = split; | |
5c680ed6 CM |
3553 | path->slots[level + 1] += 1; |
3554 | } else { | |
925baedd | 3555 | btrfs_tree_unlock(split); |
5f39d397 | 3556 | free_extent_buffer(split); |
be0e5c09 | 3557 | } |
aa5d6bed | 3558 | return ret; |
be0e5c09 CM |
3559 | } |
3560 | ||
74123bd7 CM |
3561 | /* |
3562 | * how many bytes are required to store the items in a leaf. start | |
3563 | * and nr indicate which items in the leaf to check. This totals up the | |
3564 | * space used both by the item structs and the item data | |
3565 | */ | |
5f39d397 | 3566 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3567 | { |
41be1f3b JB |
3568 | struct btrfs_item *start_item; |
3569 | struct btrfs_item *end_item; | |
3570 | struct btrfs_map_token token; | |
be0e5c09 | 3571 | int data_len; |
5f39d397 | 3572 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3573 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3574 | |
3575 | if (!nr) | |
3576 | return 0; | |
c82f823c | 3577 | btrfs_init_map_token(&token, l); |
dd3cc16b RK |
3578 | start_item = btrfs_item_nr(start); |
3579 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3580 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3581 | btrfs_token_item_size(l, start_item, &token); | |
3582 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3583 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3584 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3585 | return data_len; |
3586 | } | |
3587 | ||
d4dbff95 CM |
3588 | /* |
3589 | * The space between the end of the leaf items and | |
3590 | * the start of the leaf data. IOW, how much room | |
3591 | * the leaf has left for both items and data | |
3592 | */ | |
e902baac | 3593 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 3594 | { |
e902baac | 3595 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3596 | int nritems = btrfs_header_nritems(leaf); |
3597 | int ret; | |
0b246afa JM |
3598 | |
3599 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3600 | if (ret < 0) { |
0b246afa JM |
3601 | btrfs_crit(fs_info, |
3602 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3603 | ret, | |
3604 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3605 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3606 | } |
3607 | return ret; | |
d4dbff95 CM |
3608 | } |
3609 | ||
99d8f83c CM |
3610 | /* |
3611 | * min slot controls the lowest index we're willing to push to the | |
3612 | * right. We'll push up to and including min_slot, but no lower | |
3613 | */ | |
f72f0010 | 3614 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
3615 | int data_size, int empty, |
3616 | struct extent_buffer *right, | |
99d8f83c CM |
3617 | int free_space, u32 left_nritems, |
3618 | u32 min_slot) | |
00ec4c51 | 3619 | { |
f72f0010 | 3620 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3621 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3622 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3623 | struct btrfs_map_token token; |
5f39d397 | 3624 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3625 | int slot; |
34a38218 | 3626 | u32 i; |
00ec4c51 CM |
3627 | int push_space = 0; |
3628 | int push_items = 0; | |
0783fcfc | 3629 | struct btrfs_item *item; |
34a38218 | 3630 | u32 nr; |
7518a238 | 3631 | u32 right_nritems; |
5f39d397 | 3632 | u32 data_end; |
db94535d | 3633 | u32 this_item_size; |
00ec4c51 | 3634 | |
34a38218 CM |
3635 | if (empty) |
3636 | nr = 0; | |
3637 | else | |
99d8f83c | 3638 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3639 | |
31840ae1 | 3640 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3641 | push_space += data_size; |
31840ae1 | 3642 | |
44871b1b | 3643 | slot = path->slots[1]; |
34a38218 CM |
3644 | i = left_nritems - 1; |
3645 | while (i >= nr) { | |
dd3cc16b | 3646 | item = btrfs_item_nr(i); |
db94535d | 3647 | |
31840ae1 ZY |
3648 | if (!empty && push_items > 0) { |
3649 | if (path->slots[0] > i) | |
3650 | break; | |
3651 | if (path->slots[0] == i) { | |
e902baac DS |
3652 | int space = btrfs_leaf_free_space(left); |
3653 | ||
31840ae1 ZY |
3654 | if (space + push_space * 2 > free_space) |
3655 | break; | |
3656 | } | |
3657 | } | |
3658 | ||
00ec4c51 | 3659 | if (path->slots[0] == i) |
87b29b20 | 3660 | push_space += data_size; |
db94535d | 3661 | |
db94535d CM |
3662 | this_item_size = btrfs_item_size(left, item); |
3663 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3664 | break; |
31840ae1 | 3665 | |
00ec4c51 | 3666 | push_items++; |
db94535d | 3667 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3668 | if (i == 0) |
3669 | break; | |
3670 | i--; | |
db94535d | 3671 | } |
5f39d397 | 3672 | |
925baedd CM |
3673 | if (push_items == 0) |
3674 | goto out_unlock; | |
5f39d397 | 3675 | |
6c1500f2 | 3676 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3677 | |
00ec4c51 | 3678 | /* push left to right */ |
5f39d397 | 3679 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3680 | |
5f39d397 | 3681 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 3682 | push_space -= leaf_data_end(left); |
5f39d397 | 3683 | |
00ec4c51 | 3684 | /* make room in the right data area */ |
8f881e8c | 3685 | data_end = leaf_data_end(right); |
5f39d397 | 3686 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3687 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3688 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3689 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3690 | |
00ec4c51 | 3691 | /* copy from the left data area */ |
3d9ec8c4 | 3692 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3693 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3694 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3695 | push_space); |
5f39d397 CM |
3696 | |
3697 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3698 | btrfs_item_nr_offset(0), | |
3699 | right_nritems * sizeof(struct btrfs_item)); | |
3700 | ||
00ec4c51 | 3701 | /* copy the items from left to right */ |
5f39d397 CM |
3702 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3703 | btrfs_item_nr_offset(left_nritems - push_items), | |
3704 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3705 | |
3706 | /* update the item pointers */ | |
c82f823c | 3707 | btrfs_init_map_token(&token, right); |
7518a238 | 3708 | right_nritems += push_items; |
5f39d397 | 3709 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3710 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3711 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3712 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3713 | push_space -= btrfs_token_item_size(right, item, &token); |
3714 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3715 | } |
3716 | ||
7518a238 | 3717 | left_nritems -= push_items; |
5f39d397 | 3718 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3719 | |
34a38218 CM |
3720 | if (left_nritems) |
3721 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3722 | else |
6a884d7d | 3723 | btrfs_clean_tree_block(left); |
f0486c68 | 3724 | |
5f39d397 | 3725 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3726 | |
5f39d397 CM |
3727 | btrfs_item_key(right, &disk_key, 0); |
3728 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3729 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3730 | |
00ec4c51 | 3731 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3732 | if (path->slots[0] >= left_nritems) { |
3733 | path->slots[0] -= left_nritems; | |
925baedd | 3734 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3735 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3736 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3737 | free_extent_buffer(path->nodes[0]); |
3738 | path->nodes[0] = right; | |
00ec4c51 CM |
3739 | path->slots[1] += 1; |
3740 | } else { | |
925baedd | 3741 | btrfs_tree_unlock(right); |
5f39d397 | 3742 | free_extent_buffer(right); |
00ec4c51 CM |
3743 | } |
3744 | return 0; | |
925baedd CM |
3745 | |
3746 | out_unlock: | |
3747 | btrfs_tree_unlock(right); | |
3748 | free_extent_buffer(right); | |
3749 | return 1; | |
00ec4c51 | 3750 | } |
925baedd | 3751 | |
44871b1b CM |
3752 | /* |
3753 | * push some data in the path leaf to the right, trying to free up at | |
3754 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3755 | * | |
3756 | * returns 1 if the push failed because the other node didn't have enough | |
3757 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3758 | * |
3759 | * this will push starting from min_slot to the end of the leaf. It won't | |
3760 | * push any slot lower than min_slot | |
44871b1b CM |
3761 | */ |
3762 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3763 | *root, struct btrfs_path *path, |
3764 | int min_data_size, int data_size, | |
3765 | int empty, u32 min_slot) | |
44871b1b CM |
3766 | { |
3767 | struct extent_buffer *left = path->nodes[0]; | |
3768 | struct extent_buffer *right; | |
3769 | struct extent_buffer *upper; | |
3770 | int slot; | |
3771 | int free_space; | |
3772 | u32 left_nritems; | |
3773 | int ret; | |
3774 | ||
3775 | if (!path->nodes[1]) | |
3776 | return 1; | |
3777 | ||
3778 | slot = path->slots[1]; | |
3779 | upper = path->nodes[1]; | |
3780 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3781 | return 1; | |
3782 | ||
3783 | btrfs_assert_tree_locked(path->nodes[1]); | |
3784 | ||
4b231ae4 | 3785 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3786 | /* |
3787 | * slot + 1 is not valid or we fail to read the right node, | |
3788 | * no big deal, just return. | |
3789 | */ | |
3790 | if (IS_ERR(right)) | |
91ca338d TI |
3791 | return 1; |
3792 | ||
44871b1b | 3793 | btrfs_tree_lock(right); |
8bead258 | 3794 | btrfs_set_lock_blocking_write(right); |
44871b1b | 3795 | |
e902baac | 3796 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3797 | if (free_space < data_size) |
3798 | goto out_unlock; | |
3799 | ||
3800 | /* cow and double check */ | |
3801 | ret = btrfs_cow_block(trans, root, right, upper, | |
3802 | slot + 1, &right); | |
3803 | if (ret) | |
3804 | goto out_unlock; | |
3805 | ||
e902baac | 3806 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3807 | if (free_space < data_size) |
3808 | goto out_unlock; | |
3809 | ||
3810 | left_nritems = btrfs_header_nritems(left); | |
3811 | if (left_nritems == 0) | |
3812 | goto out_unlock; | |
3813 | ||
2ef1fed2 FDBM |
3814 | if (path->slots[0] == left_nritems && !empty) { |
3815 | /* Key greater than all keys in the leaf, right neighbor has | |
3816 | * enough room for it and we're not emptying our leaf to delete | |
3817 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3818 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3819 | btrfs_tree_unlock(left); |
3820 | free_extent_buffer(left); | |
3821 | path->nodes[0] = right; | |
3822 | path->slots[0] = 0; | |
3823 | path->slots[1]++; | |
3824 | return 0; | |
3825 | } | |
3826 | ||
f72f0010 | 3827 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3828 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3829 | out_unlock: |
3830 | btrfs_tree_unlock(right); | |
3831 | free_extent_buffer(right); | |
3832 | return 1; | |
3833 | } | |
3834 | ||
74123bd7 CM |
3835 | /* |
3836 | * push some data in the path leaf to the left, trying to free up at | |
3837 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3838 | * |
3839 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3840 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3841 | * items | |
74123bd7 | 3842 | */ |
8087c193 | 3843 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3844 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3845 | int free_space, u32 right_nritems, |
3846 | u32 max_slot) | |
be0e5c09 | 3847 | { |
8087c193 | 3848 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3849 | struct btrfs_disk_key disk_key; |
3850 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3851 | int i; |
be0e5c09 CM |
3852 | int push_space = 0; |
3853 | int push_items = 0; | |
0783fcfc | 3854 | struct btrfs_item *item; |
7518a238 | 3855 | u32 old_left_nritems; |
34a38218 | 3856 | u32 nr; |
aa5d6bed | 3857 | int ret = 0; |
db94535d CM |
3858 | u32 this_item_size; |
3859 | u32 old_left_item_size; | |
cfed81a0 CM |
3860 | struct btrfs_map_token token; |
3861 | ||
34a38218 | 3862 | if (empty) |
99d8f83c | 3863 | nr = min(right_nritems, max_slot); |
34a38218 | 3864 | else |
99d8f83c | 3865 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3866 | |
3867 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3868 | item = btrfs_item_nr(i); |
db94535d | 3869 | |
31840ae1 ZY |
3870 | if (!empty && push_items > 0) { |
3871 | if (path->slots[0] < i) | |
3872 | break; | |
3873 | if (path->slots[0] == i) { | |
e902baac DS |
3874 | int space = btrfs_leaf_free_space(right); |
3875 | ||
31840ae1 ZY |
3876 | if (space + push_space * 2 > free_space) |
3877 | break; | |
3878 | } | |
3879 | } | |
3880 | ||
be0e5c09 | 3881 | if (path->slots[0] == i) |
87b29b20 | 3882 | push_space += data_size; |
db94535d CM |
3883 | |
3884 | this_item_size = btrfs_item_size(right, item); | |
3885 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3886 | break; |
db94535d | 3887 | |
be0e5c09 | 3888 | push_items++; |
db94535d CM |
3889 | push_space += this_item_size + sizeof(*item); |
3890 | } | |
3891 | ||
be0e5c09 | 3892 | if (push_items == 0) { |
925baedd CM |
3893 | ret = 1; |
3894 | goto out; | |
be0e5c09 | 3895 | } |
fae7f21c | 3896 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3897 | |
be0e5c09 | 3898 | /* push data from right to left */ |
5f39d397 CM |
3899 | copy_extent_buffer(left, right, |
3900 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3901 | btrfs_item_nr_offset(0), | |
3902 | push_items * sizeof(struct btrfs_item)); | |
3903 | ||
0b246afa | 3904 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3905 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3906 | |
3d9ec8c4 | 3907 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3908 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3909 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3910 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3911 | push_space); |
5f39d397 | 3912 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3913 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3914 | |
c82f823c | 3915 | btrfs_init_map_token(&token, left); |
db94535d | 3916 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3917 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3918 | u32 ioff; |
db94535d | 3919 | |
dd3cc16b | 3920 | item = btrfs_item_nr(i); |
db94535d | 3921 | |
cfed81a0 CM |
3922 | ioff = btrfs_token_item_offset(left, item, &token); |
3923 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3924 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3925 | &token); |
be0e5c09 | 3926 | } |
5f39d397 | 3927 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3928 | |
3929 | /* fixup right node */ | |
31b1a2bd JL |
3930 | if (push_items > right_nritems) |
3931 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3932 | right_nritems); |
34a38218 CM |
3933 | |
3934 | if (push_items < right_nritems) { | |
3935 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 3936 | leaf_data_end(right); |
3d9ec8c4 | 3937 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3938 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3939 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3940 | leaf_data_end(right), push_space); |
34a38218 CM |
3941 | |
3942 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3943 | btrfs_item_nr_offset(push_items), |
3944 | (btrfs_header_nritems(right) - push_items) * | |
3945 | sizeof(struct btrfs_item)); | |
34a38218 | 3946 | } |
c82f823c DS |
3947 | |
3948 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3949 | right_nritems -= push_items; |
3950 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3951 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3952 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3953 | item = btrfs_item_nr(i); |
db94535d | 3954 | |
cfed81a0 CM |
3955 | push_space = push_space - btrfs_token_item_size(right, |
3956 | item, &token); | |
3957 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3958 | } |
eb60ceac | 3959 | |
5f39d397 | 3960 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3961 | if (right_nritems) |
3962 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3963 | else |
6a884d7d | 3964 | btrfs_clean_tree_block(right); |
098f59c2 | 3965 | |
5f39d397 | 3966 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3967 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3968 | |
3969 | /* then fixup the leaf pointer in the path */ | |
3970 | if (path->slots[0] < push_items) { | |
3971 | path->slots[0] += old_left_nritems; | |
925baedd | 3972 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3973 | free_extent_buffer(path->nodes[0]); |
3974 | path->nodes[0] = left; | |
be0e5c09 CM |
3975 | path->slots[1] -= 1; |
3976 | } else { | |
925baedd | 3977 | btrfs_tree_unlock(left); |
5f39d397 | 3978 | free_extent_buffer(left); |
be0e5c09 CM |
3979 | path->slots[0] -= push_items; |
3980 | } | |
eb60ceac | 3981 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3982 | return ret; |
925baedd CM |
3983 | out: |
3984 | btrfs_tree_unlock(left); | |
3985 | free_extent_buffer(left); | |
3986 | return ret; | |
be0e5c09 CM |
3987 | } |
3988 | ||
44871b1b CM |
3989 | /* |
3990 | * push some data in the path leaf to the left, trying to free up at | |
3991 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3992 | * |
3993 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3994 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3995 | * items | |
44871b1b CM |
3996 | */ |
3997 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3998 | *root, struct btrfs_path *path, int min_data_size, |
3999 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
4000 | { |
4001 | struct extent_buffer *right = path->nodes[0]; | |
4002 | struct extent_buffer *left; | |
4003 | int slot; | |
4004 | int free_space; | |
4005 | u32 right_nritems; | |
4006 | int ret = 0; | |
4007 | ||
4008 | slot = path->slots[1]; | |
4009 | if (slot == 0) | |
4010 | return 1; | |
4011 | if (!path->nodes[1]) | |
4012 | return 1; | |
4013 | ||
4014 | right_nritems = btrfs_header_nritems(right); | |
4015 | if (right_nritems == 0) | |
4016 | return 1; | |
4017 | ||
4018 | btrfs_assert_tree_locked(path->nodes[1]); | |
4019 | ||
4b231ae4 | 4020 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
4021 | /* |
4022 | * slot - 1 is not valid or we fail to read the left node, | |
4023 | * no big deal, just return. | |
4024 | */ | |
4025 | if (IS_ERR(left)) | |
91ca338d TI |
4026 | return 1; |
4027 | ||
44871b1b | 4028 | btrfs_tree_lock(left); |
8bead258 | 4029 | btrfs_set_lock_blocking_write(left); |
44871b1b | 4030 | |
e902baac | 4031 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4032 | if (free_space < data_size) { |
4033 | ret = 1; | |
4034 | goto out; | |
4035 | } | |
4036 | ||
4037 | /* cow and double check */ | |
4038 | ret = btrfs_cow_block(trans, root, left, | |
4039 | path->nodes[1], slot - 1, &left); | |
4040 | if (ret) { | |
4041 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4042 | if (ret == -ENOSPC) |
4043 | ret = 1; | |
44871b1b CM |
4044 | goto out; |
4045 | } | |
4046 | ||
e902baac | 4047 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
4048 | if (free_space < data_size) { |
4049 | ret = 1; | |
4050 | goto out; | |
4051 | } | |
4052 | ||
8087c193 | 4053 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
4054 | empty, left, free_space, right_nritems, |
4055 | max_slot); | |
44871b1b CM |
4056 | out: |
4057 | btrfs_tree_unlock(left); | |
4058 | free_extent_buffer(left); | |
4059 | return ret; | |
4060 | } | |
4061 | ||
4062 | /* | |
4063 | * split the path's leaf in two, making sure there is at least data_size | |
4064 | * available for the resulting leaf level of the path. | |
44871b1b | 4065 | */ |
143bede5 | 4066 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
4067 | struct btrfs_path *path, |
4068 | struct extent_buffer *l, | |
4069 | struct extent_buffer *right, | |
4070 | int slot, int mid, int nritems) | |
44871b1b | 4071 | { |
94f94ad9 | 4072 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
4073 | int data_copy_size; |
4074 | int rt_data_off; | |
4075 | int i; | |
44871b1b | 4076 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4077 | struct btrfs_map_token token; |
4078 | ||
44871b1b CM |
4079 | nritems = nritems - mid; |
4080 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 4081 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
4082 | |
4083 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4084 | btrfs_item_nr_offset(mid), | |
4085 | nritems * sizeof(struct btrfs_item)); | |
4086 | ||
4087 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4088 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4089 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 4090 | leaf_data_end(l), data_copy_size); |
44871b1b | 4091 | |
0b246afa | 4092 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 4093 | |
c82f823c | 4094 | btrfs_init_map_token(&token, right); |
44871b1b | 4095 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 4096 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4097 | u32 ioff; |
4098 | ||
cfed81a0 CM |
4099 | ioff = btrfs_token_item_offset(right, item, &token); |
4100 | btrfs_set_token_item_offset(right, item, | |
4101 | ioff + rt_data_off, &token); | |
44871b1b CM |
4102 | } |
4103 | ||
44871b1b | 4104 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4105 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 4106 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
4107 | |
4108 | btrfs_mark_buffer_dirty(right); | |
4109 | btrfs_mark_buffer_dirty(l); | |
4110 | BUG_ON(path->slots[0] != slot); | |
4111 | ||
44871b1b CM |
4112 | if (mid <= slot) { |
4113 | btrfs_tree_unlock(path->nodes[0]); | |
4114 | free_extent_buffer(path->nodes[0]); | |
4115 | path->nodes[0] = right; | |
4116 | path->slots[0] -= mid; | |
4117 | path->slots[1] += 1; | |
4118 | } else { | |
4119 | btrfs_tree_unlock(right); | |
4120 | free_extent_buffer(right); | |
4121 | } | |
4122 | ||
4123 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4124 | } |
4125 | ||
99d8f83c CM |
4126 | /* |
4127 | * double splits happen when we need to insert a big item in the middle | |
4128 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4129 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4130 | * A B C | |
4131 | * | |
4132 | * We avoid this by trying to push the items on either side of our target | |
4133 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4134 | * completely. | |
4135 | */ | |
4136 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4137 | struct btrfs_root *root, | |
4138 | struct btrfs_path *path, | |
4139 | int data_size) | |
4140 | { | |
4141 | int ret; | |
4142 | int progress = 0; | |
4143 | int slot; | |
4144 | u32 nritems; | |
5a4267ca | 4145 | int space_needed = data_size; |
99d8f83c CM |
4146 | |
4147 | slot = path->slots[0]; | |
5a4267ca | 4148 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 4149 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
4150 | |
4151 | /* | |
4152 | * try to push all the items after our slot into the | |
4153 | * right leaf | |
4154 | */ | |
5a4267ca | 4155 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4156 | if (ret < 0) |
4157 | return ret; | |
4158 | ||
4159 | if (ret == 0) | |
4160 | progress++; | |
4161 | ||
4162 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4163 | /* | |
4164 | * our goal is to get our slot at the start or end of a leaf. If | |
4165 | * we've done so we're done | |
4166 | */ | |
4167 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4168 | return 0; | |
4169 | ||
e902baac | 4170 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4171 | return 0; |
4172 | ||
4173 | /* try to push all the items before our slot into the next leaf */ | |
4174 | slot = path->slots[0]; | |
263d3995 FM |
4175 | space_needed = data_size; |
4176 | if (slot > 0) | |
e902baac | 4177 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 4178 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4179 | if (ret < 0) |
4180 | return ret; | |
4181 | ||
4182 | if (ret == 0) | |
4183 | progress++; | |
4184 | ||
4185 | if (progress) | |
4186 | return 0; | |
4187 | return 1; | |
4188 | } | |
4189 | ||
74123bd7 CM |
4190 | /* |
4191 | * split the path's leaf in two, making sure there is at least data_size | |
4192 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4193 | * |
4194 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4195 | */ |
e02119d5 CM |
4196 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4197 | struct btrfs_root *root, | |
310712b2 | 4198 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4199 | struct btrfs_path *path, int data_size, |
4200 | int extend) | |
be0e5c09 | 4201 | { |
5d4f98a2 | 4202 | struct btrfs_disk_key disk_key; |
5f39d397 | 4203 | struct extent_buffer *l; |
7518a238 | 4204 | u32 nritems; |
eb60ceac CM |
4205 | int mid; |
4206 | int slot; | |
5f39d397 | 4207 | struct extent_buffer *right; |
b7a0365e | 4208 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4209 | int ret = 0; |
aa5d6bed | 4210 | int wret; |
5d4f98a2 | 4211 | int split; |
cc0c5538 | 4212 | int num_doubles = 0; |
99d8f83c | 4213 | int tried_avoid_double = 0; |
aa5d6bed | 4214 | |
a5719521 YZ |
4215 | l = path->nodes[0]; |
4216 | slot = path->slots[0]; | |
4217 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4218 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4219 | return -EOVERFLOW; |
4220 | ||
40689478 | 4221 | /* first try to make some room by pushing left and right */ |
33157e05 | 4222 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4223 | int space_needed = data_size; |
4224 | ||
4225 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 4226 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4227 | |
4228 | wret = push_leaf_right(trans, root, path, space_needed, | |
4229 | space_needed, 0, 0); | |
d397712b | 4230 | if (wret < 0) |
eaee50e8 | 4231 | return wret; |
3685f791 | 4232 | if (wret) { |
263d3995 FM |
4233 | space_needed = data_size; |
4234 | if (slot > 0) | |
e902baac | 4235 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
4236 | wret = push_leaf_left(trans, root, path, space_needed, |
4237 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4238 | if (wret < 0) |
4239 | return wret; | |
4240 | } | |
4241 | l = path->nodes[0]; | |
aa5d6bed | 4242 | |
3685f791 | 4243 | /* did the pushes work? */ |
e902baac | 4244 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 4245 | return 0; |
3326d1b0 | 4246 | } |
aa5d6bed | 4247 | |
5c680ed6 | 4248 | if (!path->nodes[1]) { |
fdd99c72 | 4249 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4250 | if (ret) |
4251 | return ret; | |
4252 | } | |
cc0c5538 | 4253 | again: |
5d4f98a2 | 4254 | split = 1; |
cc0c5538 | 4255 | l = path->nodes[0]; |
eb60ceac | 4256 | slot = path->slots[0]; |
5f39d397 | 4257 | nritems = btrfs_header_nritems(l); |
d397712b | 4258 | mid = (nritems + 1) / 2; |
54aa1f4d | 4259 | |
5d4f98a2 YZ |
4260 | if (mid <= slot) { |
4261 | if (nritems == 1 || | |
4262 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4263 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4264 | if (slot >= nritems) { |
4265 | split = 0; | |
4266 | } else { | |
4267 | mid = slot; | |
4268 | if (mid != nritems && | |
4269 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4270 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4271 | if (data_size && !tried_avoid_double) |
4272 | goto push_for_double; | |
5d4f98a2 YZ |
4273 | split = 2; |
4274 | } | |
4275 | } | |
4276 | } | |
4277 | } else { | |
4278 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4279 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4280 | if (!extend && data_size && slot == 0) { |
4281 | split = 0; | |
4282 | } else if ((extend || !data_size) && slot == 0) { | |
4283 | mid = 1; | |
4284 | } else { | |
4285 | mid = slot; | |
4286 | if (mid != nritems && | |
4287 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4288 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4289 | if (data_size && !tried_avoid_double) |
4290 | goto push_for_double; | |
67871254 | 4291 | split = 2; |
5d4f98a2 YZ |
4292 | } |
4293 | } | |
4294 | } | |
4295 | } | |
4296 | ||
4297 | if (split == 0) | |
4298 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4299 | else | |
4300 | btrfs_item_key(l, &disk_key, mid); | |
4301 | ||
a6279470 FM |
4302 | right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0, |
4303 | l->start, 0); | |
f0486c68 | 4304 | if (IS_ERR(right)) |
5f39d397 | 4305 | return PTR_ERR(right); |
f0486c68 | 4306 | |
0b246afa | 4307 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4308 | |
5d4f98a2 YZ |
4309 | if (split == 0) { |
4310 | if (mid <= slot) { | |
4311 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4312 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4313 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4314 | btrfs_tree_unlock(path->nodes[0]); |
4315 | free_extent_buffer(path->nodes[0]); | |
4316 | path->nodes[0] = right; | |
4317 | path->slots[0] = 0; | |
4318 | path->slots[1] += 1; | |
4319 | } else { | |
4320 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 4321 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 4322 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
4323 | btrfs_tree_unlock(path->nodes[0]); |
4324 | free_extent_buffer(path->nodes[0]); | |
4325 | path->nodes[0] = right; | |
4326 | path->slots[0] = 0; | |
143bede5 | 4327 | if (path->slots[1] == 0) |
b167fa91 | 4328 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 4329 | } |
196e0249 LB |
4330 | /* |
4331 | * We create a new leaf 'right' for the required ins_len and | |
4332 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4333 | * the content of ins_len to 'right'. | |
4334 | */ | |
5d4f98a2 | 4335 | return ret; |
d4dbff95 | 4336 | } |
74123bd7 | 4337 | |
94f94ad9 | 4338 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 4339 | |
5d4f98a2 | 4340 | if (split == 2) { |
cc0c5538 CM |
4341 | BUG_ON(num_doubles != 0); |
4342 | num_doubles++; | |
4343 | goto again; | |
a429e513 | 4344 | } |
44871b1b | 4345 | |
143bede5 | 4346 | return 0; |
99d8f83c CM |
4347 | |
4348 | push_for_double: | |
4349 | push_for_double_split(trans, root, path, data_size); | |
4350 | tried_avoid_double = 1; | |
e902baac | 4351 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
4352 | return 0; |
4353 | goto again; | |
be0e5c09 CM |
4354 | } |
4355 | ||
ad48fd75 YZ |
4356 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4357 | struct btrfs_root *root, | |
4358 | struct btrfs_path *path, int ins_len) | |
459931ec | 4359 | { |
ad48fd75 | 4360 | struct btrfs_key key; |
459931ec | 4361 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4362 | struct btrfs_file_extent_item *fi; |
4363 | u64 extent_len = 0; | |
4364 | u32 item_size; | |
4365 | int ret; | |
459931ec CM |
4366 | |
4367 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4368 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4369 | ||
4370 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4371 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4372 | ||
e902baac | 4373 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 4374 | return 0; |
459931ec CM |
4375 | |
4376 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4377 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4378 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4379 | struct btrfs_file_extent_item); | |
4380 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4381 | } | |
b3b4aa74 | 4382 | btrfs_release_path(path); |
459931ec | 4383 | |
459931ec | 4384 | path->keep_locks = 1; |
ad48fd75 YZ |
4385 | path->search_for_split = 1; |
4386 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4387 | path->search_for_split = 0; |
a8df6fe6 FM |
4388 | if (ret > 0) |
4389 | ret = -EAGAIN; | |
ad48fd75 YZ |
4390 | if (ret < 0) |
4391 | goto err; | |
459931ec | 4392 | |
ad48fd75 YZ |
4393 | ret = -EAGAIN; |
4394 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4395 | /* if our item isn't there, return now */ |
4396 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4397 | goto err; |
4398 | ||
109f6aef | 4399 | /* the leaf has changed, it now has room. return now */ |
e902baac | 4400 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
4401 | goto err; |
4402 | ||
ad48fd75 YZ |
4403 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4404 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4405 | struct btrfs_file_extent_item); | |
4406 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4407 | goto err; | |
459931ec CM |
4408 | } |
4409 | ||
b9473439 | 4410 | btrfs_set_path_blocking(path); |
ad48fd75 | 4411 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4412 | if (ret) |
4413 | goto err; | |
459931ec | 4414 | |
ad48fd75 | 4415 | path->keep_locks = 0; |
b9473439 | 4416 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4417 | return 0; |
4418 | err: | |
4419 | path->keep_locks = 0; | |
4420 | return ret; | |
4421 | } | |
4422 | ||
25263cd7 | 4423 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 4424 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4425 | unsigned long split_offset) |
4426 | { | |
4427 | struct extent_buffer *leaf; | |
4428 | struct btrfs_item *item; | |
4429 | struct btrfs_item *new_item; | |
4430 | int slot; | |
4431 | char *buf; | |
4432 | u32 nritems; | |
4433 | u32 item_size; | |
4434 | u32 orig_offset; | |
4435 | struct btrfs_disk_key disk_key; | |
4436 | ||
b9473439 | 4437 | leaf = path->nodes[0]; |
e902baac | 4438 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4439 | |
b4ce94de CM |
4440 | btrfs_set_path_blocking(path); |
4441 | ||
dd3cc16b | 4442 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4443 | orig_offset = btrfs_item_offset(leaf, item); |
4444 | item_size = btrfs_item_size(leaf, item); | |
4445 | ||
459931ec | 4446 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4447 | if (!buf) |
4448 | return -ENOMEM; | |
4449 | ||
459931ec CM |
4450 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4451 | path->slots[0]), item_size); | |
459931ec | 4452 | |
ad48fd75 | 4453 | slot = path->slots[0] + 1; |
459931ec | 4454 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4455 | if (slot != nritems) { |
4456 | /* shift the items */ | |
4457 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4458 | btrfs_item_nr_offset(slot), |
4459 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4460 | } |
4461 | ||
4462 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4463 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4464 | ||
dd3cc16b | 4465 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4466 | |
4467 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4468 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4469 | ||
4470 | btrfs_set_item_offset(leaf, item, | |
4471 | orig_offset + item_size - split_offset); | |
4472 | btrfs_set_item_size(leaf, item, split_offset); | |
4473 | ||
4474 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4475 | ||
4476 | /* write the data for the start of the original item */ | |
4477 | write_extent_buffer(leaf, buf, | |
4478 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4479 | split_offset); | |
4480 | ||
4481 | /* write the data for the new item */ | |
4482 | write_extent_buffer(leaf, buf + split_offset, | |
4483 | btrfs_item_ptr_offset(leaf, slot), | |
4484 | item_size - split_offset); | |
4485 | btrfs_mark_buffer_dirty(leaf); | |
4486 | ||
e902baac | 4487 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 4488 | kfree(buf); |
ad48fd75 YZ |
4489 | return 0; |
4490 | } | |
4491 | ||
4492 | /* | |
4493 | * This function splits a single item into two items, | |
4494 | * giving 'new_key' to the new item and splitting the | |
4495 | * old one at split_offset (from the start of the item). | |
4496 | * | |
4497 | * The path may be released by this operation. After | |
4498 | * the split, the path is pointing to the old item. The | |
4499 | * new item is going to be in the same node as the old one. | |
4500 | * | |
4501 | * Note, the item being split must be smaller enough to live alone on | |
4502 | * a tree block with room for one extra struct btrfs_item | |
4503 | * | |
4504 | * This allows us to split the item in place, keeping a lock on the | |
4505 | * leaf the entire time. | |
4506 | */ | |
4507 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4508 | struct btrfs_root *root, | |
4509 | struct btrfs_path *path, | |
310712b2 | 4510 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4511 | unsigned long split_offset) |
4512 | { | |
4513 | int ret; | |
4514 | ret = setup_leaf_for_split(trans, root, path, | |
4515 | sizeof(struct btrfs_item)); | |
4516 | if (ret) | |
4517 | return ret; | |
4518 | ||
25263cd7 | 4519 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
4520 | return ret; |
4521 | } | |
4522 | ||
ad48fd75 YZ |
4523 | /* |
4524 | * This function duplicate a item, giving 'new_key' to the new item. | |
4525 | * It guarantees both items live in the same tree leaf and the new item | |
4526 | * is contiguous with the original item. | |
4527 | * | |
4528 | * This allows us to split file extent in place, keeping a lock on the | |
4529 | * leaf the entire time. | |
4530 | */ | |
4531 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4532 | struct btrfs_root *root, | |
4533 | struct btrfs_path *path, | |
310712b2 | 4534 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4535 | { |
4536 | struct extent_buffer *leaf; | |
4537 | int ret; | |
4538 | u32 item_size; | |
4539 | ||
4540 | leaf = path->nodes[0]; | |
4541 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4542 | ret = setup_leaf_for_split(trans, root, path, | |
4543 | item_size + sizeof(struct btrfs_item)); | |
4544 | if (ret) | |
4545 | return ret; | |
4546 | ||
4547 | path->slots[0]++; | |
afe5fea7 | 4548 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4549 | item_size, item_size + |
4550 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4551 | leaf = path->nodes[0]; |
4552 | memcpy_extent_buffer(leaf, | |
4553 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4554 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4555 | item_size); | |
4556 | return 0; | |
4557 | } | |
4558 | ||
d352ac68 CM |
4559 | /* |
4560 | * make the item pointed to by the path smaller. new_size indicates | |
4561 | * how small to make it, and from_end tells us if we just chop bytes | |
4562 | * off the end of the item or if we shift the item to chop bytes off | |
4563 | * the front. | |
4564 | */ | |
78ac4f9e | 4565 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 4566 | { |
b18c6685 | 4567 | int slot; |
5f39d397 CM |
4568 | struct extent_buffer *leaf; |
4569 | struct btrfs_item *item; | |
b18c6685 CM |
4570 | u32 nritems; |
4571 | unsigned int data_end; | |
4572 | unsigned int old_data_start; | |
4573 | unsigned int old_size; | |
4574 | unsigned int size_diff; | |
4575 | int i; | |
cfed81a0 CM |
4576 | struct btrfs_map_token token; |
4577 | ||
5f39d397 | 4578 | leaf = path->nodes[0]; |
179e29e4 CM |
4579 | slot = path->slots[0]; |
4580 | ||
4581 | old_size = btrfs_item_size_nr(leaf, slot); | |
4582 | if (old_size == new_size) | |
143bede5 | 4583 | return; |
b18c6685 | 4584 | |
5f39d397 | 4585 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4586 | data_end = leaf_data_end(leaf); |
b18c6685 | 4587 | |
5f39d397 | 4588 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4589 | |
b18c6685 CM |
4590 | size_diff = old_size - new_size; |
4591 | ||
4592 | BUG_ON(slot < 0); | |
4593 | BUG_ON(slot >= nritems); | |
4594 | ||
4595 | /* | |
4596 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4597 | */ | |
4598 | /* first correct the data pointers */ | |
c82f823c | 4599 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4600 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4601 | u32 ioff; |
dd3cc16b | 4602 | item = btrfs_item_nr(i); |
db94535d | 4603 | |
cfed81a0 CM |
4604 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4605 | btrfs_set_token_item_offset(leaf, item, | |
4606 | ioff + size_diff, &token); | |
b18c6685 | 4607 | } |
db94535d | 4608 | |
b18c6685 | 4609 | /* shift the data */ |
179e29e4 | 4610 | if (from_end) { |
3d9ec8c4 NB |
4611 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4612 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4613 | data_end, old_data_start + new_size - data_end); |
4614 | } else { | |
4615 | struct btrfs_disk_key disk_key; | |
4616 | u64 offset; | |
4617 | ||
4618 | btrfs_item_key(leaf, &disk_key, slot); | |
4619 | ||
4620 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4621 | unsigned long ptr; | |
4622 | struct btrfs_file_extent_item *fi; | |
4623 | ||
4624 | fi = btrfs_item_ptr(leaf, slot, | |
4625 | struct btrfs_file_extent_item); | |
4626 | fi = (struct btrfs_file_extent_item *)( | |
4627 | (unsigned long)fi - size_diff); | |
4628 | ||
4629 | if (btrfs_file_extent_type(leaf, fi) == | |
4630 | BTRFS_FILE_EXTENT_INLINE) { | |
4631 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4632 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4633 | (unsigned long)fi, |
7ec20afb | 4634 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4635 | } |
4636 | } | |
4637 | ||
3d9ec8c4 NB |
4638 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4639 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
4640 | data_end, old_data_start - data_end); |
4641 | ||
4642 | offset = btrfs_disk_key_offset(&disk_key); | |
4643 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4644 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4645 | if (slot == 0) | |
b167fa91 | 4646 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 4647 | } |
5f39d397 | 4648 | |
dd3cc16b | 4649 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4650 | btrfs_set_item_size(leaf, item, new_size); |
4651 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4652 | |
e902baac | 4653 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4654 | btrfs_print_leaf(leaf); |
b18c6685 | 4655 | BUG(); |
5f39d397 | 4656 | } |
b18c6685 CM |
4657 | } |
4658 | ||
d352ac68 | 4659 | /* |
8f69dbd2 | 4660 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4661 | */ |
c71dd880 | 4662 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 4663 | { |
6567e837 | 4664 | int slot; |
5f39d397 CM |
4665 | struct extent_buffer *leaf; |
4666 | struct btrfs_item *item; | |
6567e837 CM |
4667 | u32 nritems; |
4668 | unsigned int data_end; | |
4669 | unsigned int old_data; | |
4670 | unsigned int old_size; | |
4671 | int i; | |
cfed81a0 CM |
4672 | struct btrfs_map_token token; |
4673 | ||
5f39d397 | 4674 | leaf = path->nodes[0]; |
6567e837 | 4675 | |
5f39d397 | 4676 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4677 | data_end = leaf_data_end(leaf); |
6567e837 | 4678 | |
e902baac | 4679 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4680 | btrfs_print_leaf(leaf); |
6567e837 | 4681 | BUG(); |
5f39d397 | 4682 | } |
6567e837 | 4683 | slot = path->slots[0]; |
5f39d397 | 4684 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4685 | |
4686 | BUG_ON(slot < 0); | |
3326d1b0 | 4687 | if (slot >= nritems) { |
a4f78750 | 4688 | btrfs_print_leaf(leaf); |
c71dd880 | 4689 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4690 | slot, nritems); |
290342f6 | 4691 | BUG(); |
3326d1b0 | 4692 | } |
6567e837 CM |
4693 | |
4694 | /* | |
4695 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4696 | */ | |
4697 | /* first correct the data pointers */ | |
c82f823c | 4698 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4699 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4700 | u32 ioff; |
dd3cc16b | 4701 | item = btrfs_item_nr(i); |
db94535d | 4702 | |
cfed81a0 CM |
4703 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4704 | btrfs_set_token_item_offset(leaf, item, | |
4705 | ioff - data_size, &token); | |
6567e837 | 4706 | } |
5f39d397 | 4707 | |
6567e837 | 4708 | /* shift the data */ |
3d9ec8c4 NB |
4709 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4710 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4711 | data_end, old_data - data_end); |
5f39d397 | 4712 | |
6567e837 | 4713 | data_end = old_data; |
5f39d397 | 4714 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4715 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4716 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4717 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4718 | |
e902baac | 4719 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4720 | btrfs_print_leaf(leaf); |
6567e837 | 4721 | BUG(); |
5f39d397 | 4722 | } |
6567e837 CM |
4723 | } |
4724 | ||
74123bd7 | 4725 | /* |
44871b1b CM |
4726 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4727 | * to save stack depth by doing the bulk of the work in a function | |
4728 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4729 | */ |
afe5fea7 | 4730 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4731 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4732 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4733 | { |
0b246afa | 4734 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4735 | struct btrfs_item *item; |
9c58309d | 4736 | int i; |
7518a238 | 4737 | u32 nritems; |
be0e5c09 | 4738 | unsigned int data_end; |
e2fa7227 | 4739 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4740 | struct extent_buffer *leaf; |
4741 | int slot; | |
cfed81a0 CM |
4742 | struct btrfs_map_token token; |
4743 | ||
24cdc847 FM |
4744 | if (path->slots[0] == 0) { |
4745 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
b167fa91 | 4746 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4747 | } |
4748 | btrfs_unlock_up_safe(path, 1); | |
4749 | ||
5f39d397 | 4750 | leaf = path->nodes[0]; |
44871b1b | 4751 | slot = path->slots[0]; |
74123bd7 | 4752 | |
5f39d397 | 4753 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4754 | data_end = leaf_data_end(leaf); |
eb60ceac | 4755 | |
e902baac | 4756 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4757 | btrfs_print_leaf(leaf); |
0b246afa | 4758 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4759 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4760 | BUG(); |
d4dbff95 | 4761 | } |
5f39d397 | 4762 | |
c82f823c | 4763 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4764 | if (slot != nritems) { |
5f39d397 | 4765 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4766 | |
5f39d397 | 4767 | if (old_data < data_end) { |
a4f78750 | 4768 | btrfs_print_leaf(leaf); |
0b246afa | 4769 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4770 | slot, old_data, data_end); |
290342f6 | 4771 | BUG(); |
5f39d397 | 4772 | } |
be0e5c09 CM |
4773 | /* |
4774 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4775 | */ | |
4776 | /* first correct the data pointers */ | |
0783fcfc | 4777 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4778 | u32 ioff; |
db94535d | 4779 | |
62e85577 | 4780 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4781 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4782 | btrfs_set_token_item_offset(leaf, item, | |
4783 | ioff - total_data, &token); | |
0783fcfc | 4784 | } |
be0e5c09 | 4785 | /* shift the items */ |
9c58309d | 4786 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4787 | btrfs_item_nr_offset(slot), |
d6025579 | 4788 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4789 | |
4790 | /* shift the data */ | |
3d9ec8c4 NB |
4791 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4792 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4793 | data_end, old_data - data_end); |
be0e5c09 CM |
4794 | data_end = old_data; |
4795 | } | |
5f39d397 | 4796 | |
62e2749e | 4797 | /* setup the item for the new data */ |
9c58309d CM |
4798 | for (i = 0; i < nr; i++) { |
4799 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4800 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4801 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4802 | btrfs_set_token_item_offset(leaf, item, |
4803 | data_end - data_size[i], &token); | |
9c58309d | 4804 | data_end -= data_size[i]; |
cfed81a0 | 4805 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4806 | } |
44871b1b | 4807 | |
9c58309d | 4808 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4809 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4810 | |
e902baac | 4811 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4812 | btrfs_print_leaf(leaf); |
be0e5c09 | 4813 | BUG(); |
5f39d397 | 4814 | } |
44871b1b CM |
4815 | } |
4816 | ||
4817 | /* | |
4818 | * Given a key and some data, insert items into the tree. | |
4819 | * This does all the path init required, making room in the tree if needed. | |
4820 | */ | |
4821 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4822 | struct btrfs_root *root, | |
4823 | struct btrfs_path *path, | |
310712b2 | 4824 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4825 | int nr) |
4826 | { | |
44871b1b CM |
4827 | int ret = 0; |
4828 | int slot; | |
4829 | int i; | |
4830 | u32 total_size = 0; | |
4831 | u32 total_data = 0; | |
4832 | ||
4833 | for (i = 0; i < nr; i++) | |
4834 | total_data += data_size[i]; | |
4835 | ||
4836 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4837 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4838 | if (ret == 0) | |
4839 | return -EEXIST; | |
4840 | if (ret < 0) | |
143bede5 | 4841 | return ret; |
44871b1b | 4842 | |
44871b1b CM |
4843 | slot = path->slots[0]; |
4844 | BUG_ON(slot < 0); | |
4845 | ||
afe5fea7 | 4846 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4847 | total_data, total_size, nr); |
143bede5 | 4848 | return 0; |
62e2749e CM |
4849 | } |
4850 | ||
4851 | /* | |
4852 | * Given a key and some data, insert an item into the tree. | |
4853 | * This does all the path init required, making room in the tree if needed. | |
4854 | */ | |
310712b2 OS |
4855 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4856 | const struct btrfs_key *cpu_key, void *data, | |
4857 | u32 data_size) | |
62e2749e CM |
4858 | { |
4859 | int ret = 0; | |
2c90e5d6 | 4860 | struct btrfs_path *path; |
5f39d397 CM |
4861 | struct extent_buffer *leaf; |
4862 | unsigned long ptr; | |
62e2749e | 4863 | |
2c90e5d6 | 4864 | path = btrfs_alloc_path(); |
db5b493a TI |
4865 | if (!path) |
4866 | return -ENOMEM; | |
2c90e5d6 | 4867 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4868 | if (!ret) { |
5f39d397 CM |
4869 | leaf = path->nodes[0]; |
4870 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4871 | write_extent_buffer(leaf, data, ptr, data_size); | |
4872 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4873 | } |
2c90e5d6 | 4874 | btrfs_free_path(path); |
aa5d6bed | 4875 | return ret; |
be0e5c09 CM |
4876 | } |
4877 | ||
74123bd7 | 4878 | /* |
5de08d7d | 4879 | * delete the pointer from a given node. |
74123bd7 | 4880 | * |
d352ac68 CM |
4881 | * the tree should have been previously balanced so the deletion does not |
4882 | * empty a node. | |
74123bd7 | 4883 | */ |
afe5fea7 TI |
4884 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4885 | int level, int slot) | |
be0e5c09 | 4886 | { |
5f39d397 | 4887 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4888 | u32 nritems; |
f3ea38da | 4889 | int ret; |
be0e5c09 | 4890 | |
5f39d397 | 4891 | nritems = btrfs_header_nritems(parent); |
d397712b | 4892 | if (slot != nritems - 1) { |
bf1d3425 DS |
4893 | if (level) { |
4894 | ret = tree_mod_log_insert_move(parent, slot, slot + 1, | |
a446a979 | 4895 | nritems - slot - 1); |
bf1d3425 DS |
4896 | BUG_ON(ret < 0); |
4897 | } | |
5f39d397 CM |
4898 | memmove_extent_buffer(parent, |
4899 | btrfs_node_key_ptr_offset(slot), | |
4900 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4901 | sizeof(struct btrfs_key_ptr) * |
4902 | (nritems - slot - 1)); | |
57ba86c0 | 4903 | } else if (level) { |
e09c2efe DS |
4904 | ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE, |
4905 | GFP_NOFS); | |
57ba86c0 | 4906 | BUG_ON(ret < 0); |
bb803951 | 4907 | } |
f3ea38da | 4908 | |
7518a238 | 4909 | nritems--; |
5f39d397 | 4910 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4911 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4912 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4913 | /* just turn the root into a leaf and break */ |
5f39d397 | 4914 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4915 | } else if (slot == 0) { |
5f39d397 CM |
4916 | struct btrfs_disk_key disk_key; |
4917 | ||
4918 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4919 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4920 | } |
d6025579 | 4921 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4922 | } |
4923 | ||
323ac95b CM |
4924 | /* |
4925 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4926 | * path->nodes[1]. |
323ac95b CM |
4927 | * |
4928 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4929 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4930 | * | |
4931 | * The path must have already been setup for deleting the leaf, including | |
4932 | * all the proper balancing. path->nodes[1] must be locked. | |
4933 | */ | |
143bede5 JM |
4934 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4935 | struct btrfs_root *root, | |
4936 | struct btrfs_path *path, | |
4937 | struct extent_buffer *leaf) | |
323ac95b | 4938 | { |
5d4f98a2 | 4939 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4940 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4941 | |
4d081c41 CM |
4942 | /* |
4943 | * btrfs_free_extent is expensive, we want to make sure we | |
4944 | * aren't holding any locks when we call it | |
4945 | */ | |
4946 | btrfs_unlock_up_safe(path, 0); | |
4947 | ||
f0486c68 YZ |
4948 | root_sub_used(root, leaf->len); |
4949 | ||
3083ee2e | 4950 | extent_buffer_get(leaf); |
5581a51a | 4951 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4952 | free_extent_buffer_stale(leaf); |
323ac95b | 4953 | } |
74123bd7 CM |
4954 | /* |
4955 | * delete the item at the leaf level in path. If that empties | |
4956 | * the leaf, remove it from the tree | |
4957 | */ | |
85e21bac CM |
4958 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4959 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4960 | { |
0b246afa | 4961 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4962 | struct extent_buffer *leaf; |
4963 | struct btrfs_item *item; | |
ce0eac2a AM |
4964 | u32 last_off; |
4965 | u32 dsize = 0; | |
aa5d6bed CM |
4966 | int ret = 0; |
4967 | int wret; | |
85e21bac | 4968 | int i; |
7518a238 | 4969 | u32 nritems; |
be0e5c09 | 4970 | |
5f39d397 | 4971 | leaf = path->nodes[0]; |
85e21bac CM |
4972 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4973 | ||
4974 | for (i = 0; i < nr; i++) | |
4975 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4976 | ||
5f39d397 | 4977 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4978 | |
85e21bac | 4979 | if (slot + nr != nritems) { |
8f881e8c | 4980 | int data_end = leaf_data_end(leaf); |
c82f823c | 4981 | struct btrfs_map_token token; |
5f39d397 | 4982 | |
3d9ec8c4 | 4983 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4984 | data_end + dsize, |
3d9ec8c4 | 4985 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4986 | last_off - data_end); |
5f39d397 | 4987 | |
c82f823c | 4988 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4989 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4990 | u32 ioff; |
db94535d | 4991 | |
dd3cc16b | 4992 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4993 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4994 | btrfs_set_token_item_offset(leaf, item, | |
4995 | ioff + dsize, &token); | |
0783fcfc | 4996 | } |
db94535d | 4997 | |
5f39d397 | 4998 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4999 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5000 | sizeof(struct btrfs_item) * |
85e21bac | 5001 | (nritems - slot - nr)); |
be0e5c09 | 5002 | } |
85e21bac CM |
5003 | btrfs_set_header_nritems(leaf, nritems - nr); |
5004 | nritems -= nr; | |
5f39d397 | 5005 | |
74123bd7 | 5006 | /* delete the leaf if we've emptied it */ |
7518a238 | 5007 | if (nritems == 0) { |
5f39d397 CM |
5008 | if (leaf == root->node) { |
5009 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5010 | } else { |
f0486c68 | 5011 | btrfs_set_path_blocking(path); |
6a884d7d | 5012 | btrfs_clean_tree_block(leaf); |
143bede5 | 5013 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5014 | } |
be0e5c09 | 5015 | } else { |
7518a238 | 5016 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5017 | if (slot == 0) { |
5f39d397 CM |
5018 | struct btrfs_disk_key disk_key; |
5019 | ||
5020 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 5021 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 5022 | } |
aa5d6bed | 5023 | |
74123bd7 | 5024 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5025 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5026 | /* push_leaf_left fixes the path. |
5027 | * make sure the path still points to our leaf | |
5028 | * for possible call to del_ptr below | |
5029 | */ | |
4920c9ac | 5030 | slot = path->slots[1]; |
5f39d397 CM |
5031 | extent_buffer_get(leaf); |
5032 | ||
b9473439 | 5033 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5034 | wret = push_leaf_left(trans, root, path, 1, 1, |
5035 | 1, (u32)-1); | |
54aa1f4d | 5036 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5037 | ret = wret; |
5f39d397 CM |
5038 | |
5039 | if (path->nodes[0] == leaf && | |
5040 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5041 | wret = push_leaf_right(trans, root, path, 1, |
5042 | 1, 1, 0); | |
54aa1f4d | 5043 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5044 | ret = wret; |
5045 | } | |
5f39d397 CM |
5046 | |
5047 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5048 | path->slots[1] = slot; |
143bede5 | 5049 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5050 | free_extent_buffer(leaf); |
143bede5 | 5051 | ret = 0; |
5de08d7d | 5052 | } else { |
925baedd CM |
5053 | /* if we're still in the path, make sure |
5054 | * we're dirty. Otherwise, one of the | |
5055 | * push_leaf functions must have already | |
5056 | * dirtied this buffer | |
5057 | */ | |
5058 | if (path->nodes[0] == leaf) | |
5059 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5060 | free_extent_buffer(leaf); |
be0e5c09 | 5061 | } |
d5719762 | 5062 | } else { |
5f39d397 | 5063 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5064 | } |
5065 | } | |
aa5d6bed | 5066 | return ret; |
be0e5c09 CM |
5067 | } |
5068 | ||
7bb86316 | 5069 | /* |
925baedd | 5070 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5071 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5072 | * returns < 0 on io errors. | |
d352ac68 CM |
5073 | * |
5074 | * This may release the path, and so you may lose any locks held at the | |
5075 | * time you call it. | |
7bb86316 | 5076 | */ |
16e7549f | 5077 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5078 | { |
925baedd CM |
5079 | struct btrfs_key key; |
5080 | struct btrfs_disk_key found_key; | |
5081 | int ret; | |
7bb86316 | 5082 | |
925baedd | 5083 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5084 | |
e8b0d724 | 5085 | if (key.offset > 0) { |
925baedd | 5086 | key.offset--; |
e8b0d724 | 5087 | } else if (key.type > 0) { |
925baedd | 5088 | key.type--; |
e8b0d724 FDBM |
5089 | key.offset = (u64)-1; |
5090 | } else if (key.objectid > 0) { | |
925baedd | 5091 | key.objectid--; |
e8b0d724 FDBM |
5092 | key.type = (u8)-1; |
5093 | key.offset = (u64)-1; | |
5094 | } else { | |
925baedd | 5095 | return 1; |
e8b0d724 | 5096 | } |
7bb86316 | 5097 | |
b3b4aa74 | 5098 | btrfs_release_path(path); |
925baedd CM |
5099 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5100 | if (ret < 0) | |
5101 | return ret; | |
5102 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5103 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5104 | /* |
5105 | * We might have had an item with the previous key in the tree right | |
5106 | * before we released our path. And after we released our path, that | |
5107 | * item might have been pushed to the first slot (0) of the leaf we | |
5108 | * were holding due to a tree balance. Alternatively, an item with the | |
5109 | * previous key can exist as the only element of a leaf (big fat item). | |
5110 | * Therefore account for these 2 cases, so that our callers (like | |
5111 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5112 | * the previous key we computed above. | |
5113 | */ | |
5114 | if (ret <= 0) | |
925baedd CM |
5115 | return 0; |
5116 | return 1; | |
7bb86316 CM |
5117 | } |
5118 | ||
3f157a2f CM |
5119 | /* |
5120 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5121 | * for nodes or leaves that are have a minimum transaction id. |
5122 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5123 | * |
5124 | * This does not cow, but it does stuff the starting key it finds back | |
5125 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5126 | * key and get a writable path. | |
5127 | * | |
3f157a2f CM |
5128 | * This honors path->lowest_level to prevent descent past a given level |
5129 | * of the tree. | |
5130 | * | |
d352ac68 CM |
5131 | * min_trans indicates the oldest transaction that you are interested |
5132 | * in walking through. Any nodes or leaves older than min_trans are | |
5133 | * skipped over (without reading them). | |
5134 | * | |
3f157a2f CM |
5135 | * returns zero if something useful was found, < 0 on error and 1 if there |
5136 | * was nothing in the tree that matched the search criteria. | |
5137 | */ | |
5138 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5139 | struct btrfs_path *path, |
3f157a2f CM |
5140 | u64 min_trans) |
5141 | { | |
5142 | struct extent_buffer *cur; | |
5143 | struct btrfs_key found_key; | |
5144 | int slot; | |
9652480b | 5145 | int sret; |
3f157a2f CM |
5146 | u32 nritems; |
5147 | int level; | |
5148 | int ret = 1; | |
f98de9b9 | 5149 | int keep_locks = path->keep_locks; |
3f157a2f | 5150 | |
f98de9b9 | 5151 | path->keep_locks = 1; |
3f157a2f | 5152 | again: |
bd681513 | 5153 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5154 | level = btrfs_header_level(cur); |
e02119d5 | 5155 | WARN_ON(path->nodes[level]); |
3f157a2f | 5156 | path->nodes[level] = cur; |
bd681513 | 5157 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5158 | |
5159 | if (btrfs_header_generation(cur) < min_trans) { | |
5160 | ret = 1; | |
5161 | goto out; | |
5162 | } | |
d397712b | 5163 | while (1) { |
3f157a2f CM |
5164 | nritems = btrfs_header_nritems(cur); |
5165 | level = btrfs_header_level(cur); | |
a74b35ec | 5166 | sret = btrfs_bin_search(cur, min_key, level, &slot); |
cbca7d59 FM |
5167 | if (sret < 0) { |
5168 | ret = sret; | |
5169 | goto out; | |
5170 | } | |
3f157a2f | 5171 | |
323ac95b CM |
5172 | /* at the lowest level, we're done, setup the path and exit */ |
5173 | if (level == path->lowest_level) { | |
e02119d5 CM |
5174 | if (slot >= nritems) |
5175 | goto find_next_key; | |
3f157a2f CM |
5176 | ret = 0; |
5177 | path->slots[level] = slot; | |
5178 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5179 | goto out; | |
5180 | } | |
9652480b Y |
5181 | if (sret && slot > 0) |
5182 | slot--; | |
3f157a2f | 5183 | /* |
de78b51a ES |
5184 | * check this node pointer against the min_trans parameters. |
5185 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5186 | */ |
d397712b | 5187 | while (slot < nritems) { |
3f157a2f | 5188 | u64 gen; |
e02119d5 | 5189 | |
3f157a2f CM |
5190 | gen = btrfs_node_ptr_generation(cur, slot); |
5191 | if (gen < min_trans) { | |
5192 | slot++; | |
5193 | continue; | |
5194 | } | |
de78b51a | 5195 | break; |
3f157a2f | 5196 | } |
e02119d5 | 5197 | find_next_key: |
3f157a2f CM |
5198 | /* |
5199 | * we didn't find a candidate key in this node, walk forward | |
5200 | * and find another one | |
5201 | */ | |
5202 | if (slot >= nritems) { | |
e02119d5 | 5203 | path->slots[level] = slot; |
b4ce94de | 5204 | btrfs_set_path_blocking(path); |
e02119d5 | 5205 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5206 | min_trans); |
e02119d5 | 5207 | if (sret == 0) { |
b3b4aa74 | 5208 | btrfs_release_path(path); |
3f157a2f CM |
5209 | goto again; |
5210 | } else { | |
5211 | goto out; | |
5212 | } | |
5213 | } | |
5214 | /* save our key for returning back */ | |
5215 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5216 | path->slots[level] = slot; | |
5217 | if (level == path->lowest_level) { | |
5218 | ret = 0; | |
3f157a2f CM |
5219 | goto out; |
5220 | } | |
b4ce94de | 5221 | btrfs_set_path_blocking(path); |
4b231ae4 | 5222 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
5223 | if (IS_ERR(cur)) { |
5224 | ret = PTR_ERR(cur); | |
5225 | goto out; | |
5226 | } | |
3f157a2f | 5227 | |
bd681513 | 5228 | btrfs_tree_read_lock(cur); |
b4ce94de | 5229 | |
bd681513 | 5230 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5231 | path->nodes[level - 1] = cur; |
f7c79f30 | 5232 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5233 | } |
5234 | out: | |
f98de9b9 FM |
5235 | path->keep_locks = keep_locks; |
5236 | if (ret == 0) { | |
5237 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5238 | btrfs_set_path_blocking(path); | |
3f157a2f | 5239 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5240 | } |
3f157a2f CM |
5241 | return ret; |
5242 | } | |
5243 | ||
5244 | /* | |
5245 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5246 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5247 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5248 | * |
5249 | * 0 is returned if another key is found, < 0 if there are any errors | |
5250 | * and 1 is returned if there are no higher keys in the tree | |
5251 | * | |
5252 | * path->keep_locks should be set to 1 on the search made before | |
5253 | * calling this function. | |
5254 | */ | |
e7a84565 | 5255 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5256 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5257 | { |
e7a84565 CM |
5258 | int slot; |
5259 | struct extent_buffer *c; | |
5260 | ||
6a9fb468 | 5261 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 5262 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5263 | if (!path->nodes[level]) |
5264 | return 1; | |
5265 | ||
5266 | slot = path->slots[level] + 1; | |
5267 | c = path->nodes[level]; | |
3f157a2f | 5268 | next: |
e7a84565 | 5269 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5270 | int ret; |
5271 | int orig_lowest; | |
5272 | struct btrfs_key cur_key; | |
5273 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5274 | !path->nodes[level + 1]) | |
e7a84565 | 5275 | return 1; |
33c66f43 | 5276 | |
6a9fb468 | 5277 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
5278 | level++; |
5279 | continue; | |
5280 | } | |
5281 | ||
5282 | slot = btrfs_header_nritems(c) - 1; | |
5283 | if (level == 0) | |
5284 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5285 | else | |
5286 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5287 | ||
5288 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5289 | btrfs_release_path(path); |
33c66f43 YZ |
5290 | path->lowest_level = level; |
5291 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5292 | 0, 0); | |
5293 | path->lowest_level = orig_lowest; | |
5294 | if (ret < 0) | |
5295 | return ret; | |
5296 | ||
5297 | c = path->nodes[level]; | |
5298 | slot = path->slots[level]; | |
5299 | if (ret == 0) | |
5300 | slot++; | |
5301 | goto next; | |
e7a84565 | 5302 | } |
33c66f43 | 5303 | |
e7a84565 CM |
5304 | if (level == 0) |
5305 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5306 | else { |
3f157a2f CM |
5307 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5308 | ||
3f157a2f CM |
5309 | if (gen < min_trans) { |
5310 | slot++; | |
5311 | goto next; | |
5312 | } | |
e7a84565 | 5313 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5314 | } |
e7a84565 CM |
5315 | return 0; |
5316 | } | |
5317 | return 1; | |
5318 | } | |
5319 | ||
97571fd0 | 5320 | /* |
925baedd | 5321 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5322 | * returns 0 if it found something or 1 if there are no greater leaves. |
5323 | * returns < 0 on io errors. | |
97571fd0 | 5324 | */ |
234b63a0 | 5325 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5326 | { |
5327 | return btrfs_next_old_leaf(root, path, 0); | |
5328 | } | |
5329 | ||
5330 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5331 | u64 time_seq) | |
d97e63b6 CM |
5332 | { |
5333 | int slot; | |
8e73f275 | 5334 | int level; |
5f39d397 | 5335 | struct extent_buffer *c; |
8e73f275 | 5336 | struct extent_buffer *next; |
925baedd CM |
5337 | struct btrfs_key key; |
5338 | u32 nritems; | |
5339 | int ret; | |
8e73f275 | 5340 | int old_spinning = path->leave_spinning; |
bd681513 | 5341 | int next_rw_lock = 0; |
925baedd CM |
5342 | |
5343 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5344 | if (nritems == 0) |
925baedd | 5345 | return 1; |
925baedd | 5346 | |
8e73f275 CM |
5347 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5348 | again: | |
5349 | level = 1; | |
5350 | next = NULL; | |
bd681513 | 5351 | next_rw_lock = 0; |
b3b4aa74 | 5352 | btrfs_release_path(path); |
8e73f275 | 5353 | |
a2135011 | 5354 | path->keep_locks = 1; |
31533fb2 | 5355 | path->leave_spinning = 1; |
8e73f275 | 5356 | |
3d7806ec JS |
5357 | if (time_seq) |
5358 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5359 | else | |
5360 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5361 | path->keep_locks = 0; |
5362 | ||
5363 | if (ret < 0) | |
5364 | return ret; | |
5365 | ||
a2135011 | 5366 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5367 | /* |
5368 | * by releasing the path above we dropped all our locks. A balance | |
5369 | * could have added more items next to the key that used to be | |
5370 | * at the very end of the block. So, check again here and | |
5371 | * advance the path if there are now more items available. | |
5372 | */ | |
a2135011 | 5373 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5374 | if (ret == 0) |
5375 | path->slots[0]++; | |
8e73f275 | 5376 | ret = 0; |
925baedd CM |
5377 | goto done; |
5378 | } | |
0b43e04f LB |
5379 | /* |
5380 | * So the above check misses one case: | |
5381 | * - after releasing the path above, someone has removed the item that | |
5382 | * used to be at the very end of the block, and balance between leafs | |
5383 | * gets another one with bigger key.offset to replace it. | |
5384 | * | |
5385 | * This one should be returned as well, or we can get leaf corruption | |
5386 | * later(esp. in __btrfs_drop_extents()). | |
5387 | * | |
5388 | * And a bit more explanation about this check, | |
5389 | * with ret > 0, the key isn't found, the path points to the slot | |
5390 | * where it should be inserted, so the path->slots[0] item must be the | |
5391 | * bigger one. | |
5392 | */ | |
5393 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5394 | ret = 0; | |
5395 | goto done; | |
5396 | } | |
d97e63b6 | 5397 | |
d397712b | 5398 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5399 | if (!path->nodes[level]) { |
5400 | ret = 1; | |
5401 | goto done; | |
5402 | } | |
5f39d397 | 5403 | |
d97e63b6 CM |
5404 | slot = path->slots[level] + 1; |
5405 | c = path->nodes[level]; | |
5f39d397 | 5406 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5407 | level++; |
8e73f275 CM |
5408 | if (level == BTRFS_MAX_LEVEL) { |
5409 | ret = 1; | |
5410 | goto done; | |
5411 | } | |
d97e63b6 CM |
5412 | continue; |
5413 | } | |
5f39d397 | 5414 | |
925baedd | 5415 | if (next) { |
bd681513 | 5416 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5417 | free_extent_buffer(next); |
925baedd | 5418 | } |
5f39d397 | 5419 | |
8e73f275 | 5420 | next = c; |
bd681513 | 5421 | next_rw_lock = path->locks[level]; |
d07b8528 | 5422 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5423 | slot, &key); |
8e73f275 CM |
5424 | if (ret == -EAGAIN) |
5425 | goto again; | |
5f39d397 | 5426 | |
76a05b35 | 5427 | if (ret < 0) { |
b3b4aa74 | 5428 | btrfs_release_path(path); |
76a05b35 CM |
5429 | goto done; |
5430 | } | |
5431 | ||
5cd57b2c | 5432 | if (!path->skip_locking) { |
bd681513 | 5433 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5434 | if (!ret && time_seq) { |
5435 | /* | |
5436 | * If we don't get the lock, we may be racing | |
5437 | * with push_leaf_left, holding that lock while | |
5438 | * itself waiting for the leaf we've currently | |
5439 | * locked. To solve this situation, we give up | |
5440 | * on our lock and cycle. | |
5441 | */ | |
cf538830 | 5442 | free_extent_buffer(next); |
d42244a0 JS |
5443 | btrfs_release_path(path); |
5444 | cond_resched(); | |
5445 | goto again; | |
5446 | } | |
8e73f275 CM |
5447 | if (!ret) { |
5448 | btrfs_set_path_blocking(path); | |
bd681513 | 5449 | btrfs_tree_read_lock(next); |
8e73f275 | 5450 | } |
31533fb2 | 5451 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5452 | } |
d97e63b6 CM |
5453 | break; |
5454 | } | |
5455 | path->slots[level] = slot; | |
d397712b | 5456 | while (1) { |
d97e63b6 CM |
5457 | level--; |
5458 | c = path->nodes[level]; | |
925baedd | 5459 | if (path->locks[level]) |
bd681513 | 5460 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5461 | |
5f39d397 | 5462 | free_extent_buffer(c); |
d97e63b6 CM |
5463 | path->nodes[level] = next; |
5464 | path->slots[level] = 0; | |
a74a4b97 | 5465 | if (!path->skip_locking) |
bd681513 | 5466 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5467 | if (!level) |
5468 | break; | |
b4ce94de | 5469 | |
d07b8528 | 5470 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5471 | 0, &key); |
8e73f275 CM |
5472 | if (ret == -EAGAIN) |
5473 | goto again; | |
5474 | ||
76a05b35 | 5475 | if (ret < 0) { |
b3b4aa74 | 5476 | btrfs_release_path(path); |
76a05b35 CM |
5477 | goto done; |
5478 | } | |
5479 | ||
5cd57b2c | 5480 | if (!path->skip_locking) { |
bd681513 | 5481 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5482 | if (!ret) { |
5483 | btrfs_set_path_blocking(path); | |
bd681513 | 5484 | btrfs_tree_read_lock(next); |
bd681513 | 5485 | } |
31533fb2 | 5486 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5487 | } |
d97e63b6 | 5488 | } |
8e73f275 | 5489 | ret = 0; |
925baedd | 5490 | done: |
f7c79f30 | 5491 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5492 | path->leave_spinning = old_spinning; |
5493 | if (!old_spinning) | |
5494 | btrfs_set_path_blocking(path); | |
5495 | ||
5496 | return ret; | |
d97e63b6 | 5497 | } |
0b86a832 | 5498 | |
3f157a2f CM |
5499 | /* |
5500 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5501 | * searching until it gets past min_objectid or finds an item of 'type' | |
5502 | * | |
5503 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5504 | */ | |
0b86a832 CM |
5505 | int btrfs_previous_item(struct btrfs_root *root, |
5506 | struct btrfs_path *path, u64 min_objectid, | |
5507 | int type) | |
5508 | { | |
5509 | struct btrfs_key found_key; | |
5510 | struct extent_buffer *leaf; | |
e02119d5 | 5511 | u32 nritems; |
0b86a832 CM |
5512 | int ret; |
5513 | ||
d397712b | 5514 | while (1) { |
0b86a832 | 5515 | if (path->slots[0] == 0) { |
b4ce94de | 5516 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5517 | ret = btrfs_prev_leaf(root, path); |
5518 | if (ret != 0) | |
5519 | return ret; | |
5520 | } else { | |
5521 | path->slots[0]--; | |
5522 | } | |
5523 | leaf = path->nodes[0]; | |
e02119d5 CM |
5524 | nritems = btrfs_header_nritems(leaf); |
5525 | if (nritems == 0) | |
5526 | return 1; | |
5527 | if (path->slots[0] == nritems) | |
5528 | path->slots[0]--; | |
5529 | ||
0b86a832 | 5530 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5531 | if (found_key.objectid < min_objectid) |
5532 | break; | |
0a4eefbb YZ |
5533 | if (found_key.type == type) |
5534 | return 0; | |
e02119d5 CM |
5535 | if (found_key.objectid == min_objectid && |
5536 | found_key.type < type) | |
5537 | break; | |
0b86a832 CM |
5538 | } |
5539 | return 1; | |
5540 | } | |
ade2e0b3 WS |
5541 | |
5542 | /* | |
5543 | * search in extent tree to find a previous Metadata/Data extent item with | |
5544 | * min objecitd. | |
5545 | * | |
5546 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5547 | */ | |
5548 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5549 | struct btrfs_path *path, u64 min_objectid) | |
5550 | { | |
5551 | struct btrfs_key found_key; | |
5552 | struct extent_buffer *leaf; | |
5553 | u32 nritems; | |
5554 | int ret; | |
5555 | ||
5556 | while (1) { | |
5557 | if (path->slots[0] == 0) { | |
5558 | btrfs_set_path_blocking(path); | |
5559 | ret = btrfs_prev_leaf(root, path); | |
5560 | if (ret != 0) | |
5561 | return ret; | |
5562 | } else { | |
5563 | path->slots[0]--; | |
5564 | } | |
5565 | leaf = path->nodes[0]; | |
5566 | nritems = btrfs_header_nritems(leaf); | |
5567 | if (nritems == 0) | |
5568 | return 1; | |
5569 | if (path->slots[0] == nritems) | |
5570 | path->slots[0]--; | |
5571 | ||
5572 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5573 | if (found_key.objectid < min_objectid) | |
5574 | break; | |
5575 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5576 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5577 | return 0; | |
5578 | if (found_key.objectid == min_objectid && | |
5579 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5580 | break; | |
5581 | } | |
5582 | return 1; | |
5583 | } |